1

Demographics and Entrepreneurship

James Liang*

Hui Wang**

Edward P. Lazear***

Version: December 2014

*Guanghua School of Management, Peking University

**Guanghua School of Management, Peking University

***Graduate School of Business and Hoover Institution, Stanford University.

Lazear gratefully acknowledges support of the Coulter Faculty Fellowship.

JEL code: J11, L26, M51

Keywords: demographics, entrepreneurship, within firm age structure, seniority effect

2

Abstract

Entrepreneurship requires creativity and business acumen. Creativity may

decline with age, but business skills increase with experience in high level

positions. Having too many older workers in society slows entrepreneurship.

Not only are older workers less innovative, but more significant is that when

older workers occupy key positions they block younger workers from

acquiring business skills. A formal theoretical structure is presented and

tested using the Global Entrepreneurship Monitor data. The results imply that

a one-standard deviation decrease in the median age of a country increases the

rate of new business formation by 2.5 percentage points, which is about forty

percent of the mean rate. Furthermore, older societies have lower rates of

entrepreneurship at every age.

3

1. Introduction

Few doubt the role of entrepreneurship in fostering economic growth. Without new

business creation, it is difficult to imagine that game-changing technologies can make their

way into the economic environment. Existing companies can modernize and update their

products and techniques of production, but the major innovations tend to be associated with

entrepreneurship and the formation of new companies.

Many significant inventions of the last 150 years illustrate the point. Thomas Edison

invented the light bulb and founded General Electric. The inventor of the automobile was

Karl Friedrich Benz, followed closely by Gottlieb Wilhelm Daimler. Daimler-Benz is the

product of their inventions. Alexander Graham Bell invented the telephone and founded

AT&T. Guglielmo Marconi, the inventor of radio, was a founder of Wireless Telegraph &

Signal in Britain. The Wright Brothers founded The Wright Company, which later became

Curtiss-Wright. Steven Wosniak, who invented the personal computer, teamed up with Steve

Jobs to form Apple. The list goes on.

In the last 50 years, most of the developed world has experienced a dramatic

demographic shift. Japan is a case in point. Almost immediately after World War II, Japan’s

fertility rate fell rapidly and dropped below replacement level in the 1960s. Currently,

Japan’s fertility rate is one of the lowest in the world at 1.3, which means each generation

will shrink by 40%. Similarly, the fertility rates of the European countries also fell in the

50-60s, and currently are at around 1.6 on average, which means each generation will shrink

by 20%. In fact, the U.S. is the only major developed country with replacement level fertility

and some major developing countries, like China, also have below replacement fertility rates.

In the near future, an aging and shrinking workforce likely will be the norm for most of the

world.

The general trend towards an aging and shrinking workforce is unprecedented. Its impact

on future economic development can be profound. Standard economic theory views the effect

4

of aging on the economy primarily through its effect on reducing the size of the labor force,

and through the fiscal drain associated with government support of a larger fraction of the

population. This effect can be offset at least partially by extending the retirement age, as

many countries have already done.1,2

The focus here is on another mechanism by which not only the supply of workforce, but

also the age structure of the workforce, can have a significant impact on economic

performance through the channel of entrepreneurship. Some studies already attribute Japan’s

"lost decades" to its "entrepreneurship vacuum" since the 1990s. But what causes this

vacuum? Age structure is the answer emphasized here. A worker in a country with a

younger workforce, like the United States, will be more entrepreneurial than a worker in a

country with an older workforce, like Japan.

The argument is that entrepreneurial capability depends on two types of abilities:

creativity and business acumen. Although the results offer no direct test of either of the effect

of creativity or business acumen on entrepreneurship, the findings are in complete

alignment with the predictions of the theory based on these two factors. Furthermore, some

of the borne-out predictions are subtle and are not produced in a coherent fashion by other

hypotheses.

Creativity refers to the ability to think in novel ways so as to break away from products

and production methods of the past. In the theory, younger workers are assumed to be more

creative. Additionally, business skill is acquired through on-the-job experience, which

depends on the jobs to which individuals are assigned. Workers who do low-level, menial

tasks are unlikely to acquire the kind of business skill that will make them successful

entrepreneurs. Workers who are placed in roles that give them decision making authority

1 Schwarz and Demirgue-Kunt (1999) document 17 countries around the world that extended their mandatory

retirement age during 1992 to 1998. Many of these are OECD countries, e.g., Czech Republic (change from 60

to 62), Ireland (65 to 66), Italy (60 to 62), and New Zealand (60 to 65). Examples occurring after 1998 include

France, changing from 60 to 62 in 2010, and Germany, changing from 65 to 67 in 2012. 2 In fact, many scholars hold the view that the threat of aging has been overstated. Proposed solutions to deal

with the labor shortage problem include: extension of retirement age (Herrmann, 2012), attracting international

migration from labor-surplus countries (Bloom, Canning, and Fink, 2010, Herrmann, 2012), accumulation of

human capital and increases in labor productivity (Prettner, 2013), etc. A recent paper by Vogel, Ludwig, and

Borsch-Supan (2013) evaluates these alternative methods and finds that policies promoting human capital

formation in combination with an increase in the retirement age are effective in reducing welfare loss due to

demographic transitions.

5

and the experience of different management situations acquire more business skills that

translate into operating a successful firm.

The skill acquisition factor, which derives from Becker’s seminal work on human capital

(e.g., Becker, 1962, 1975), accounts for the bulk of the findings. Becker postulated that

workers would acquire human capital through on-the-job training and that the augmentation

of their stock of human capital would affect their productivity and earning capacity. That

same view is adopted here. Workers may begin with raw talent and inherent creativity, but

the acquisition of skills at work is essential to their founding a business. It is for that reason

that the young are not the ones most likely to start businesses, even if they are the most

creative. They must have time to obtain the skills on the job that will allow business that they

found to succeed.

Becker (1960) emphasized the ability of economics to explain fertility and demographic

structure. Becker showed how prices, the most important of which was the implicit price of

time as captured by the wage rate of the mother, affected desired fertility. This analysis

takes the demographic structure as given, but allows demographics to affect the rate of skill

acquisition. It is this interaction between demographics and skill acquisition that allows the

economic framework below, which is in the spirit of Becker, to explain how entrepreneurship

rates vary across countries, by age, and over time.

As an extension of the Becker notion of human capital accumulation on the job, it is

hypothesized that rank in the firm affects an individual’s exposure to experiences that

produce the human capital necessary to start a business. Rank proxies the opportunity to

encounter the relevant experiences. Workers in higher positions shoulder more responsibility,

have more interaction with other decision makers and are in positions to see the larger picture.

The higher one is in an organization, the more opportunity to gain experience that will be

useful in starting an enterprise.

It is for this reason that the demographic structure of a country affects human capital

formation. The probability of being in top positions depends on the age structure of the

workforce. If a firm has an old workforce, it is less likely that the younger workers will be

6

given much management opportunity because the slots are already occupied by the more

senior workers. In young firms, even top positions are held by relatively young people. It is

for that reason that the age structure of a country is potentially an important determinant of

entrepreneurship. A young society provides more opportunity for the young and most

creative to acquire the skills necessary for entrepreneurship. Combining the notion that the

young may be more creative with the Becker idea that skills are acquired on the job, over

time, and only when the environment for learning is right, generates two important

predictions.3

First, the relation of entrepreneurship to age is inverted u-shaped. The very young do not

have the requisite human capital and the very old have lost their creativity. The second

prediction links demographics directly to entrepreneurship. Workers are less likely to become

entrepreneurs in a country where the cohort size is shrinking over time. In an aging country,

there is a higher proportion of senior workers, which slows down promotion of junior

workers. As a consequence, human capital accumulates more slowly for the younger workers

because they must wait longer to be in those positions that are most conducive to skill

production. As a result, in steady-state, workers at every age have less of the human capital

required to start businesses and entrepreneurship is suppressed. This mechanism is labeled

the "rank effect," and is new to the literature.

The model is tested using a unique dataset on cross-country entrepreneurship called the

Global Entrepreneurship Monitor (Global Entrepreneurship Research Association, GERA,

2010). It is, at this point, the only large-scale cross-country dataset on entrepreneurial activity

and age. The cross-country regressions support the theory laid out below and the effects of

demographics are large. The estimates imply that a median age that is one standard deviation

3 See Galenson (2001). Acemoglu, Akcigit and Celik (2014) find that innovation is related to a society’s

openness to disruptive ideas. They find that manager and inventor age are negatively related to the

innovativeness of an invention. The general willingness to accept disruption may also be inversely related to

the age of a society. The specific implications that are derived by our theory are not part of their argument.

For example, the inverted u-shape relation of entrepreneurship with age and the fact that younger countries have

a less pronounced u-shape pattern, derived below, come directly from the theory in this paper, but not from

other theories absent additional modifications.

7

lower is associated with a 2.5 percentage point higher country rate of entrepreneurship, which

is about 40% of the mean rate. This effect is significant both statistically and economically,

and is robust across different specifications, alternative measures of entrepreneurship, and

among OECD and non-OECD countries.

To better understand this point estimate, it is important to distinguish the "rank" from the

simple "creativity" explanation, the latter merely saying that young people are more

innovative. Most directly, without the rank effect or some other considerations4

,

entrepreneurship would fall monotonically with age. The youngest individuals in the labor

force would be the ones starting businesses. In fact, that is not the case. Although there is

variation across countries, Figure 1 (discussed in more detail below) makes clear that the

likelihood of starting a business rises at least until the 30s and in some countries, later in life.

To explain this, it is necessary that other attributes necessary for business creation rise with

age.

An alternative would be to simply assume that the relation of creativity to age is

inverted-u-shaped. The second piece of evidence provides the strongest support for the rank

effect. Within every age group, the entrepreneurship rate is lower in countries that are older.

This can have nothing to do with entrepreneurship rates that vary between ages, but must

instead be explained by something that relates entrepreneurship at a given age to the

demographic structure of the country. Although there may be other factors that are captured

by a country’s demographic structure that could affect entrepreneurship, it will be shown that

the specific predictions of the theory that links human capital acquisition to the age

distribution is completely consistent with the findings in ways that other views are not.

The data used provide no direct evidence on creativity and rank (or skill acquisition)

effects.5 However, as mentioned, the model gives very specific predictions, all of which are

supported by the data. Of course, there may be alternative mechanisms that could yield

4 For example, the ability to obtain physical capital for investment might be age related.

5 In a typical seniority based firm, with a larger older cohort, young workers are promoted much more slowly.

In 1976, 32% of the "Manager" grade workers were younger than 35; in 1994 the ratio dropped to only 16%

(Dore, 1996). Consistent with the human capital acquisition notion, the age earnings profile has flattened. For

workers born from 1945 to 1949, real wages at age 35 are about 2.6 times the starting wage. For workers born

20 years later, real wages at age 35 are only 2.0 times the starting wage (Shire, 2008).

8

similar implications. An attempt is made to address some alternative theories, which do not

seem to be supported by the results. The theory presented below is logical, straightforward,

and supported by the evidence.

As an empirical matter, the rank effect is more important than the composition effect.

Although the young may be more creative than the old, the fact that the young are denied

training opportunities afforded by higher rank accounts for the low rate of entrepreneurship in

an aging society. Cross-country within-age group variation does the bulk of the work, not

the fact that the young are more entrepreneurial than the old in the average country.

2. The Effect of Age on Entrepreneurship

2.1. A Motivating Anecdote: Japan’s Entrepreneur Vacuum and the Lost Decades

The Japanese economy grew rapidly in the 1970 and 1980s, and many economists

predicted that it would soon over-take the U.S. in terms of GDP per capita. However, after

the bust of real estate bubble in 1991, the Japanese economy stagnated during the following

20 years, while the U.S. economy pulled ahead again, benefiting from a vigorous high-tech

industry. Hot debates about the cause of Japan’s lost decades are still going on, but one

possible explanation of slow growth is the lack of entrepreneurship and the failure to develop

a dynamic IT industry as the U.S. did during the last 30 years. Five of the top 10 high tech

companies in the U.S. were founded after 1985, and their founders were also very young

when they established these companies, with an average age of only 28. By contrast, in Japan,

none of the top 10 high-tech companies were founded in the last 40 years. New firm entry

rate dropped from the 6 to 7% range in the 1960s and 1970s to 3% in the 1990s (Acht,

Thunik, and Verheut, 2004), which amounts to less than 1/3 of that in the U.S. and trails all

the other OECD countries (Karlin, 2013). According to the entrepreneurship survey used in

this paper, entrepreneurial propensity in Japan is the lowest among all the developed

countries. In the US, 4.9% adults between the ages of 18 and 64 are working actively to

establish new businesses, as compared to only 1.9% in Japan.

9

Interestingly, it should be noted that Japan’s “entrepreneur vacuum” is not due to the lack

of technological investment. The country’s R&D spending accounts for about 3% of its GDP,

ahead of all the other OECD countries. Japan still enjoys considerable patent advantages in

Asia (Karlin, 2013). However, many of these patents are significantly underutilized. Perhaps

the cause of Japan’s lagging in entrepreneurship is not the dearth of ideas or opportunities,

but the dearth of necessary capabilities to recognize and exploit these opportunities and to

convert the innovative ideas into creation of new firms.

2.2. Determinants of Entrepreneurship over Lifetime

Among all the determinants of the decision to engage in entrepreneurship, age is among

the most obvious factors to consider (see Parker, 2006, p106). Data from the Global

Entrepreneurship Monitor, Figure 1, which divides countries into young, middle, and old,

show the changes in entrepreneurial propensity at different ages. One noticeable stylized fact

is that all three curves are inverted U-shaped, with the likelihood of becoming entrepreneur

peaking for the young-middle aged people. This observation is consistent with many existing

studies (e.g., Blanchflower and Meyer, 1994; Blanchflower and Oswald, 2009;

Modragon-Velez, 2009).

[Figure 1 is here]

Some attributes necessary for successful entrepreneurship clearly decrease over the life

cycle. As a logical matter, the productivity-age pattern for virtually all work activities must

be an inverted-u. Young children do not have the skills, strength or wisdom to be very

productive. The very old possess neither the energy, stamina, nor mental acuity to carry out

complex assignments. The issue is determining the peak of the U and how the shape of the

productivity-age relationship varies with the activity in question. There is much research,

primarily outside economics, on age-productivity patterns. For example, Ruth and Birren

(1985) discuss the decline in the ability to conduct logical thinking and reasoning as an

individual ages. Florida (2002) argues that creativity can wane with aging. The young have

advantages in the ability to store and process information, solve problems, deal with

10

complexity, and adjust to new situations (Kaufman and Horn, 1996; Ryan, Sattler, and Lopez,

2000). A recent paper by Acemoglu et al.(2014) studies the relationship between creativity

and age of the inventor and the manager, and find that inventor and manager age are

negatively correlated with creativeness of the invention.

Still, successful entrepreneurship must require more than the energy and creativity of

youth or the entrepreneurship rate would peak among fresh graduates (around 20 years old).

Lazear (2004a) found that entrepreneurship demands a diverse set of skills, rather than

specialized knowledge. Many of these skills need to be accumulated through work, social

interactions, and learning-by-doing. For example, “tacit knowledge” accumulated over a

lifetime peaks when a person is in his or her 50s (Wang and Kaufman 1993; Kaufman and

Horn 1996; Ryan, Sattler, and Lopez, 2000). Moreover, it is helpful for aspiring

entrepreneurs to develop a strong social network, which is assists their ability to access

resources in different phrase of establishment process (Granovetter, 1985). The importance of

networks in launching a business has been well documented by empirical studies in different

cultural settings (e.g, Greve and Salaff, 2003; Wellman, 1999).

2.3. Demographic Structure and Entrepreneurship

Another pattern apparent in Figure 1 is that entrepreneurship in countries with a younger

workforce first-order dominates those in countries with an older workforce. At every age, the

entrepreneurship rate of the older countries is lower than that of middle and younger

countries. This implies that the propensity to become an entrepreneur at any given age is

lower when the country has a higher median age. Furthermore, the effect is especially

pronounced for middle-aged individuals between 30 and 40 years of age. Both empirical

findings, that older countries have lower entrepreneurship at every age and that the

differences are greatest for middle age workers, are direct implications of the model below

(and do not follow easily from alternative explanations).

3. Model

11

3.1. Age structure of the Workforce

Suppose a country’s cohort size by age is given by

ra

re

e

rNraf

1),( 0

(1)

where N0 is the country’s population size, a is age, and r is the shrinkage parameter that

relates the size of one cohort to another.6 Normalize age a such that a=0 is age 20, i.e. which

might be a typical starting work age, and a=1 is age 65, i.e., a typical retirement age. Then

the fraction of workforce under age a, or the age distribution c.d.f.,7 is given by the following

expression:

1

1),(

r

ra

e

eraF (2)

Figure 2 compares the age distribution of a balanced workforce (r=0) to the age

distributions of shrinking (r=0.22) and growing workforces (r=-2.18). (These two values of r

are chosen to reflect the empirical parameter estimates for Japan and Uganda, respectively,

which are at the two extremes of shrinking and growing populations.)8

[Figure 2 is here]

Two properties of the population c.d.f. are important for deriving the propositions that

follow later.

First,

Lemma 1. For 1

1),(

r

ra

e

eraF , There exists an ]1,0[ma such that for all maa ,

02

drda

Fd, and for maa , 0

2

drda

Fd.

Proof: See appendix.

6 This exponential form in age is a simplified version of the conventional function that captures the stable age

distribution. See Lotka (1922). 7 The corresponding formula for the c.d.f. is:

1

00),(/),(),( daraSdaraSraF

a

8 The estimation of r is discussed below in section 5.1.

12

Lemma 1 says that the rank of the middle-aged workers shifts more with a reduction in

the population growth rate than the rank of the very young and very old.

Second, and more important for the logic of the propositions that follow later, is lemma

2.

Lemma 2. For 1

1

r

ra

e

eF and 0<a<1, 0

dr

dF

Proof: See appendix.

This states that the proportion of people below age a decreases in r for all ages (0 and 1

excepted, of course). This is shown in figure 2(b) as a downward shift in the c.d.f. for as r

increases.

3.2. The Value of Starting a Business

To model business formation, think of each person getting a random draw of an idea for

a new product or method of production in each period. The novelty of the idea affects its

market value. An idea is not sufficient, though, to create a successful business. In order to be

successful, an idea must be implemented and the probability of successful implementation

depends on the skills that a worker has acquired. Some luck is involved in getting a good idea,

but skill and creativity affect the ability to come up with a significant innovation.

The expected value of an idea combines luck, ability and youth as follows. Let v denote

the expected value of an idea relative to a worker’s current wage. Then

v =V(h, q) ξ (3)

where ξ is a random variable that captures luck, h reflects business skills and q captures

creativity so that V1>0 and V2>0.

Assume that ξ is distributed over the range (1,∞). Or equivalently, 1/ξ is distributed over

the range (0,1) with a c.d.f. P(1/ξ). The risk-neutral worker will choose to start a business if

v > 1 or if V(h,q) > 1/ξ (4)

13

The probability of starting a business is therefore P(V(h,q)).Although it is unnecessary to

specify the nature of the ξ distribution, the general structure allows for a wide range of

possibilities. For example, one possibility is to allow9

ξ = A + ut (5)

where A is a specific person’s ability (like IQ), which remains constant over time, and ut is

luck of that the individual encounters in period t in getting an idea. At one extreme, the

variance in A could be zero, in which case there would be independence of business

formation across people over time. History of entrepreneurship would say nothing about the

likelihood of starting a new business. At the other extreme, were the variance ut equal to zero,

then there would be serial entrepreneurship, where a person who started a business in one

period would be the one who starts a business in subsequent periods. All of the variation in

entrepreneurship would be across people, with none over time.10

Other than luck, there are two factors that affect the expected value of an idea. The first,

q, is creativity and is determined by the function q=Q(a). Creativity is assumed to decrease

with age, Q'(a) < 0.The second factor, h, reflects the stock of business skills, which is a

reflection of past acquisitions of human capital. As argued earlier, a worker’s ability to

acquire skills depends on his position in the firm, which depends on his rank, defined as the

proportion of workers younger than he. Let s denote rank. Then, h=H(s) with H’>0. If

everyone is younger than someone of age a, then that individual has had the most opportunity

to acquire the skills necessary to start a business. Conversely, if everyone is older than

someone of age a, other cohorts have occupied the positions in the firm that provide the

experience most valuable for entrepreneurship.

Assuming that the age distributions of every firm are the same as that of the general

population, then s≡F(a,r). The intuition is clear. Within each age group, the larger is the

proportion of individuals who are younger than a worker, or equivalently, the smaller the

proportion who are older than a worker, the more likely is that worker to hold positions that

provide him with business skills. Also, lemma 2 implies that if r0 < r1 then F(a, r0) > F(a, r1)

9 This transitory and permanent ability structure was used in Lazear (2004b).

10 See LaFontaine and Shaw (2014).

14

∀a. This implies that the worker who has more opportunity to gain business skills today also

had more opportunity to gain those skills in the past because fewer are older than every

cohort with r0 than with r1 .

3.3. Entrepreneurial Activity

Recalling equation (4), a risk-neutral individual11

will start a business if

V(h,q) > 1/ξ

with V1, V2 > 0. So the fraction of the workers at age a that start a business (denoted as E(a,r))

is given by:

)))()),,((((),( aQraFHVPraE (6)

where P(x) is the probability that x>1/ξ .

Equation (6) provides the basis of the theoretical propositions and empirical implications,

which are tested below. It says that the entrepreneurship rate for any cohort a depends on the

age distribution of the population as summarized by r, the shrinkage rate of the population.

The primary results to be tested and the empirical motivation for the analysis follow.

Proposition 1. Entrepreneurship at any given age a is decreasing in the population parameter,

r. As r rises, reflecting both a declining and aging population, entrepreneurship falls.

Specifically,

0),(

r

raE

(7)

Proof: This follows directly from (6) since

21'),(

FHVPr

raE

with P’ V1H’ > 0 and F2 < 0 from (1).

11

Alternatively, if we assume a constant relative risk aversion, the results in the model are essentially

unchanged. Moreover, we can assume risk aversion increases with age. The effect is similar to having a steeper

declining Q function, which will further reduce entrepreneurship of the older group.

15

Entrepreneurial activity decreases for every age group when r increases. An increase in r

tilts the age distribution in favor of older workers. This corresponds to the fact, discussed in

the introduction for selected countries, that the entrepreneurship propensity is lower at every

given age for country with higher r.

Next, define sa as the share of the population in a country that is below age a so sa =

F(a,r) (the i subscript for country i is suppressed). Then

Corollary 1. For any given age group, a, the entrepreneurship rate rises in sa or

0),(

a

a

s

saG .

Proof: Noting that sa = F(a,r), rewrite (6) as

G(a,sa) = P(V(H(sa),Q(a))) (6')

Then

0''),(

1

HVP

s

saG

a

a

Propositions 1 and Corollary 1 relate to the rank effect. In aging societies, a worker at

every age has a smaller proportion of the workforce below him. Because more slowly

growing populations are disproportionately older, the ability of each age group to acquire the

business skills necessary to become an entrepreneur is reduced. Consequently, the rate of

business creation slows for every given age group.

Another corollary also follows.

Corollary 2. 0),(

a

saG a

Proof: 0''),(

2

QVP

a

saG a because Q’<0 .

16

Corollary 2 says that older age groups should have lower entrepreneurship rates for any

given share of the population that is below the group. Put simply, thirty-year-olds in country

A should have lower rates of entrepreneurship than 29-year-olds in country B when s30 in

country A equals s29 in country B. (Note that this does not contradict the earlier point that

entrepreneurship rates first rise and then fall in age. In that comparison, sa is not held

constant.)

Proposition 1 and its corollaries are not sufficient to guarantee that the rate of

entrepreneurship for the country as a whole declines with r. To make that statement, it must

also be true that aging of the workforce does not shift the age distribution (sufficiently) in the

direction of those age groups that are most likely to engage in entrepreneurial activity. Were

this to happen, it is possible that the entrepreneurship rate for the country as a whole could

rise from the change in composition of age groups, despite the fact that within every age

group, entrepreneurship rates declined.

To derive the effect of changes in population growth on overall entrepreneurship, first

integrate equation (6) over ages to obtain the total number of entrepreneurs as a fraction of

the workforce in a country

1

0),(),()( radFraErE

(8)

Then the following proposition relates the country- (average-) level of entrepreneurship to

population growth.

Proposition 2. The number of entrepreneurs as a fraction of the workforce decreases as the

population ages, i.e., 0/ drEd .

Proof. See Appendix A.

Proposition 2 says that the average rate of entrepreneurship for the country as a whole

declines as the population shrinks and ages. The expected rate of entrepreneurship varies with

population growth rates both through the combination of the rank effect and the composition

17

effect of weighting groups with higher rates of entrepreneurship differently as the population

parameter, r, changes. The latter is the standard composition effect that influences all

averages across groups. It also yields the following corollary, which can be used in the

empirical analysis.

Corollary 3. The number of entrepreneurs as a fraction of the workforce decreases with the

population’s median age.

Proof: See appendix.

Additionally, the entrepreneurial activity of the middle-aged group will be affected the

most by a reduction in the population growth rate. By lemma 1, the rank distribution shifts

down the most in the middle. It is also true that the entrepreneurship rate shifts down the

most in the middle, as shown in Figure 1and formalized by Proposition 3.

First note that entrepreneurship varies with age and under certain conditions, peaks in

middle age. From (6)

))((),(

211 QVFHVPa

raE

(9)

Since P’, V1H’F1, and P2 are all positive and Q’ <0 , the sign of ∂E/∂a is indeterminate. This

is as it should be because it is expected that the propensity to start businesses first rises with

age and then declines.

More concretely, sufficient conditions can be specified that guarantee that E(a,r) first

rises in a and then declines. The conditions are, first, that creativity does not decline much

with age for the youngest individuals, specifically,

0)('lim0

aQa

and second, that the increment to business skills declines to zero as rank goes to one

0)('lim1

sHs

18

The logic behind the latter condition is that once a cohort is older than most of the

population, it already has sufficient seniority to hold the top jobs. At that point, additional

seniority has little value for entrepreneurship training.

Given that Q’(0)=0,

0))((),0(

11

FHVP

a

rE

Also, since s=1 when a=1,

0))((),1(

2

QVP

a

rE

Thus, entrepreneurship first rises in age and then eventually declines in age. Thus, it is

possible to state the following proposition.

Proposition 3

Given the assumptions that

0)('lim0

aQa

and that

0)('lim1

sHs

the entrepreneurship rate rises in age initially and then declines in age at the end of life.

Furthermore, there exists an age am with 0 < am < 1 such that E(am , r ) > E(a, r) ∀ a ≠ am .

Proof: It has already been shown that 0),0(

a

rE, which means that there exists some am >

0 for which E(am ,r ) > E(0, r) . Similarly, because 0),1(

a

rE, it is also true that

E(am,r) < E(1,r) . Simply define am to be that value of a that maximizes the rate of

entrepreneurship.

19

Proposition 3 implies that there is a peak age of entrepreneurship somewhere in the

middle of life.12

Very young workers are less entrepreneurial than those at peak

entrepreneurial age because they have not acquired the business skills. Very old workers

are less entrepreneurial than peak age because they have lost their creative edge. The inverted

u-shape relation of entrepreneurship to age distinguishes the theory set out here from other

possible explanations of cross-country correlations between entrepreneurship and population

demographics, as discussed in section 5, below.

The effect of demographics on the entrepreneurship rates is greater for some middle ages

than it is for very old and very young ages. Specifically, it is certain that there is some age

greater than zero and less than one at which the effect of r on E is more negative than it is at

either ages of 0 or 1. This is formalized in the following proposition.

Proposition 4. There exists some aM, 0<aM<1 such that r

rE

r

raE M

),0(),( and

r

rE

r

raE M

),1(

),( under the conditions that the first and second derivatives of V, H, Q exist

and are bounded and that the first derivatives of V and H are greater than some small constant

ε.

Proof. See Appendix A.

This proposition says that entrepreneurship rates of the middle-aged workers are the ones

most sensitive to cross-country demographic differences.13

In aging countries, it is the

middle aged workers, as defined by age equal to aM, who are most prone to decrease their

rates of entrepreneurship as the population ages. Again, as spelled out in section 5, other

theories that link entrepreneurship to demographics do not provide that implication without

additional ad hoc assumptions.

12

Without additional assumptions, it cannot be guaranteed that there is only one peak age. It is certain, however,

that there exists some a* between 0 and 1 at which E(a*,r) > E(a,r) ∀ a ≠a* . 13

Note that entrepreneurship declines in r so Proposition 4 says that the effect gets closer to zero at the two

extreme ages of zero and one.

20

3.4.Country Level Aggregation and Summary of Empirical Predictions

Proposition 1 and its corollaries give the key predictions to be examined using the

cross-country data. These theoretical statements imply that the rate of entrepreneurship within

any age group is directly related to the rate of population growth and that holding constant the

share of the population below any age group, older groups have lower rates of

entrepreneurship than younger groups.

At the country level, demographics affect aggregate entrepreneurship through two

channels – composition and rank. The effect of composition works through Q(a). By

assumption, younger workers are more creative. The proportion of the workforce that is

young is determined by F(a,r) so increases in r imply a smaller proportion of young workers,

which means less overall creativity in the population.

The effect of rank works through H(F(a,r)). An older workforce prevents young workers

from obtaining the learning opportunities that help them start businesses. This skill deficiency

stays with them throughout their careers relative to the stock of skills that they would have at

any age had they acquired them when young. Aging populations generate low transmission of

business skills to the young, creative members of society.

The empirical implications are clear and follow directly from the theory. They are:

1. Within any country, the effect of age on entrepreneurship is negative, holding the

share of those below that age group constant. This is the implication of Corollary 2.

2. Holding age constant, the higher is sa the higher is the rate of entrepreneurship. Stated

differently, within any given age group, countries that have higher proportions of the

population below that age should have higher rates of entrepreneurship. This is the

implication of Corollary 1.

3.Countries that are aging more quickly, captured by higher levels of r, should have

lower rates of entrepreneurship at any given age. This follows from Proposition 1.

4. Categorizing countries by their r, those with higher values of r should have lower rates

of entrepreneurship overall. This is Proposition 2.

21

5. It also follows that countries with higher median ages should have lower

entrepreneurship rates. This is the result of corollary 3.

6. Within a country, entrepreneurship rates rise with age and then decline after some

point. This holds under the assumptions 0)('lim0

aQa

and

0)('lim1

sHs

. This is the result of

Proposition 3.

7. The entrepreneurship rates of the middle-aged are most sensitive to cross-country

changes in r. This follows from Proposition 4.

4. Data

4.1 Measures of Entrepreneurship

The primary data source for the cross-country comparisons is the Global

Entrepreneurship Monitor (GEM), collected by a not-for-profit company, Global

Entrepreneurship Research Association (GERA), in UK. Starting from 2000, GERA

conducted telephone and face-to-face individual level interviews with representative sample

of adult sample across a wide range of countries, evaluating their entrepreneurial activities,

aspirations, and attitudes. The dataset has several features that suit the purpose of this study.

First, unlike other datasets that measure entrepreneurship at the firm level, GEM dataset

captures entrepreneurial choices and decisions at the individual level, which is consistent

with the theoretical modeling framework. Second, GEM has a wide range of coverage,

including both developed countries and developing ones.14

The 2000-2010 GEM dataset

used contains more than 1.3 million individuals aged between 15 and 60 in 82 countries. It

provides considerable cross-country variation in the age distribution, which is helpful for

testing the hypothesis that demographic structure affects entrepreneurship. Third, GEM

questionnaires are designed under a uniform framework with the goal of producing data that

are comparable across countries.

14

GEM is an ambitious project, with an estimated global budget of nearly $9 million . Its 2013 survey is

designed to cover 75% of world population and 89% of world GDP (Bosma et al., 2012).

22

There are a number of different entrepreneurship rates that are reported in the GEM.

For most of the empirical analysis, entrepreneurship is defined as “manages and owns a

business that is up to 42 months old and pays wages”. For robustness, section 5.6 reports

results based on three alternative definitions of entrepreneurship. The first alternate definition,

“all new businesses” includes businesses that have died in the last twelve months in addition

to those already included by the 42 month criterion. Although the start date of a business that

failed is not reported, it is well-known that new business are much more likely to fail than old

ones so the deaths are probably dominated by newly formed firms. The second definition

looks only at startups, defined as an owner actively involved in a startup effort that is not

currently paying wages. The final definition, “total early stage entrepreneurship,” includes

startups that do not yet pay wages in addition to those managed and owned that are up to 42

months old. Businesses that failed are omitted from this definition. The qualitative

conclusions are not sensitive to the choice of definition of entrepreneurship.

GEM also provides information on the aspiration levels of the entrepreneurs. First, the

questionnaire asks about the number of employees expected to be hired in the next five years

for the early-stage entrepreneurial activities. We construct two variables showing

entrepreneurs if they intend to hire more than five (75 percentile) and more than 10 (90

percentile) worker, as two alternative measures of entrepreneurs with high aspiration. Second,

GEM also asks the respondents to state whether the newly-created business that they are

running involves new products or services. This is supposed to capture those entrepreneurs

with innovative ambition.

Other than entrepreneurial activities, basic demographic information such as age of the

interviewees is recorded as well. This study focuses on the population aged 20 to 64.

The survey was carried out in different countries in different years from 2001 to 2010.

Many countries were surveyed multiple times, but the panel is an unbalanced one. (Refer to

table B1 in Appendix B for detailed information on countries, timing of the survey, and

sample size in the GEM data.) The total number of observations used at the individual level,

which is confined to those between 20 and 64 years of age, inclusive, is 1.3 million. The

23

number of country-year cells is 393, and the number of country-age-year cells is 16,661. The

entrepreneurship rate can be constructed at each cell level.

4.2 Demographics

The population statistics come from the U.S. Census Bureau’s International Data Base

(IDB).15

The IDB is updated routinely and contains estimates and projections for over 200

countries and areas of the world. The IDB provides population counts by age from age 0 to

100-plus for each of the countries every year. The age-specific population detail contained in

the IDB data is used by GEM as one important source to calculate its sampling weights

(Bosma et al., 2012, page 56). Every country in the GEM sample from 2001 to 2010 can be

matched with information from the IDB dataset to construct sa,it, which is the share of the

population below any age a in country i in year t .

4.3 Other Country Characteristics

Four country level attributes serve as controls in some of the analyses. First, year- and

country-specific GDP-per-capita is obtained from Penn World Table, version 7.1 (Heston,

Summers, and Aten, 2012). The data are PPP adjusted and deflated at the 2005 price level to

make the numbers comparable across countries and over time. Second, the country- and

year-specific rates of completed tertiary education among the adult population are obtained

from a newly constructed education attainment dataset by Barro and Lee (2010). Third, year-

and country-specific “start-up costs,” defined as the cost to register a business as a

percentage of gross national income (GNI) per capita in 2010, is available from the World

Bank database.16

Fourth, an international property rights index (IPRI) of is obtained from

Property Rights Alliance (Strokova, 2010). It is constructed based on Legal and Political

Environment, Physical Property Rights (PPR), and the Intellectual Property Rights (IPR)

situation in each country. The index ranges between 0 and 10 and serves as a barometer of the

security of property rights. This variable is available for all countries in the sample only in

15

Available at: http://www.census.gov/population/international/data/idb/index.php 16

Available at http://data.worldbank.org/

24

2010. As a result, the same value is assumed to hold in that country in all years. Last, we

obtain the length of compulsory military service for each country in our sample from

Wikipedia.17

Table 1 presents the summary statistics (in 2010) of the main variables used in

this paper.

[Table 1 is here]

5. Empirical Implementation

The theory provides specific implications about what should be observed empirically. In

what follows, those theoretical implications are tested using the Global Entrepreneurship

Monitor data.

Equation (6) implies that the entrepreneurship rate of each age group is affected by

population growth parameters and in particular, by the age distribution of the population. The

age distribution of the population is observable for each country, as is the entrepreneurship

rate.

There are two variables of interest. The first is the age of the particular group. Younger

individuals are hypothesized to be more creative, which works through Q(a) directly.

Additionally, the rank effect says that for any given age group, a, the higher the rank s of that

group in the overall population, the higher is the entrepreneurship rate because those

individuals have had better opportunities to acquire the skills necessary for entrepreneurship.

This works through H(s), or equivalently, through H(F(a,r)). As described in the data section,

both a and s are observable in GEM and Population census data, respectively, a being the

normalized age, which is a number between zero and one, and s being the proportion of

individuals in a country who have ages younger than a.

The basic approach then is to estimate a model that captures the essence of equation (6).

This requires examining the cross-country differences within an age group as well as

differences within the “typical” country as age varies.

For most analyses, the unit of analysis is the country-age group because there is no

variation in the independent variables at levels of aggregation more fine than that. Because

17 Available at: http://en.wikipedia.org/wiki/Military_service. Accessed in Dec, 2014.

25

age and the proportion of the population below that age are the key variables, and because

they do not vary at the individual level, the proper level of aggregation for most purposes is

the age-country cell.18

5.1. Summarizing the Demographic Structure

The model of the demographic structure is simple where one parameter, r, captures the

entire demographic structure (see equation (1)). This structure is not necessary because the

actual sa from the population data rather than the predicted ones as dictated by equation (1)

can be used to create the explanatory variables. There is no need to rely on a single

population statistic because the data pertinent to each country-age cell are readily available.

Still, it is useful for two reasons to see whether the demographic model postulated in (1)

does a reasonable job of fitting the data. First, if it does, then the theoretical predictions of the

model that are stated in terms of r are more applicable to the actual countries studied. Second,

comparative statics, which are intuitive and straightforward in terms of the single parameter r,

or almost equivalently as will be shown, median age, have clear interpretations.

There are a variety of ways to show the goodness-of-fit of the r model. For any given

year, in each country, r can be estimated using the 45 age groups for that particular country.

Using the c.d.f. form of equation (2), r is fitted for each country and each year taken

separately, using non-linear least squares (NLS), where the dependent variable is the actual

value of sa for each age group within that country and the independent variable is simply a,

with r to be estimated. There are 45 observations for each country corresponding to each age

group within the country in a given year.

For the purposes of testing goodness-of-fit, the data from 2010 are used. Non-linear

estimation generates r for each country. Given r , a predicted sa can be obtained for each

of the 45 age groups in each country as

1

1ˆ

ˆ

ˆ

r

ar

ae

es (10)

18

All estimates of standard errors are based on clustering at the country level, which is the level of treatment, in

order to take into account that errors might not be independent, especially given the relatively fine partition of

age. Additionally, standard errors are derived after weighting for cell size.

26

Then the actual sa is regressed on the predicted as using the 57 countries included in the 2010

GEM data. There are 2,561 observations.19

The regression generates an r-square of .998,

which means that the share of the population below any given age a is predicted almost

perfectly for the 2,561 age-country cells by the form where the age is the actual age

(normalized) of the group in question and r is the country-specific r. Thus, the model of

equation (1) and (2) fits the actual demographic data very well.20

Additionally, r is an almost perfect predictor of the country’s median age. The

country-size weighted correlation between r and median age across the 57 countries in 2010

is 0.98. As a consequence, representing the country’s age structure by the more commonly

used median age instead of r is likely to accurate for most purposes.

5.2. Country level analysis

It is first instructive to examine the raw data at the country level. Figure 3(a) plots a

country’s entrepreneurship rate against the median age (among those 20 to 64) of a country in

2010. Figure 3(b) is the same as Figure 3(a), but replaces median age with the country

estimated r. The figures show 57 points, one for each country. The clear pattern is that the

older the country’s population, the lower is the entrepreneurship rate.

[Figure 3 is here]

The raw data appear to support proposition 2, which states that higher r (older) countries

should have lower entrepreneurship rates. The various panels plot the entrepreneurship rate

against r and against median age. They appear the same, which is not surprising since r and

median age are so highly correlated. The pattern holds across all countries and also just

within the OECD countries in panels 3(c) and 3(d), although the statistical relation for OECD

countries in the single year of 2010 alone, which is what is shown, is not statistically

significant.

19

In principle, there should be (45 age groups) x (57 countries) = 2,565 age-country cells. But four of these

cells with age above 60 are empty, which leaves us 2,561 cells. 20

Much of this is because sa is highly correlated with a. Even if cross-country variation is omitted and the

mean value of r is used to predict sa , the r-squared is .96. Despite the correlation, sa and a have independent

and important effects on entrepreneurship rates, as shown below.

27

As a starting point, Table 2 reports the results of a more systematic analysis, presenting

results of reduced form regressions at the country level of aggregation. Each observation in

this table is a country-year combination, as available in the GEM data. To account for the fact

that different countries were surveyed in different years and some countries were surveyed

more years than other countries, year dummies are included in all specifications. Standard

errors are clustered at the country level.21

[Table 2 is here]

Columns 1 and 2 regress the country’s entrepreneurship rate on r, as estimated above.

Column 1 does this for all countries, whereas column 2 restricts the analysis to OECD

countries only. The results are clear - the higher the value of r, the lower the entrepreneurship

rate. This is true in the full sample and also in the OECD restricted sample, albeit somewhat

weaker in the OECD sample.

Columns 3 and 4 repeat the analysis, but replace r with the median age (among 20-64

year-olds) in the country. The results are similar.

Columns 5 and 6 add other explanatory variables. They are: the log of per-capita GDP,

the five-year per-capita average growth rate, the tertiary education completion rate, an

estimate of start-up costs as a percentage of gross national income, an index of intellectual

property and a dummy for whether the country has more than one year of compulsory

military service.22

Although one or more of these variables sometimes enter significantly

(depending on the subsample used), it remains true that r continues to be an important

explanatory variable. The demographic structure does not appear to be a proxy for the other

potential explanatory factors, at least the obvious candidates for which data are available.

This is important because it suggests that other interpretations of demographics, like the

average age of the population proxying the growth potential of the country, do not find strong

support in the data. The more refined analyses below add more credence to this result.

21

Counry-year clustering was also done, when relevant. None of the results on the demograhic variables was

affected in any subtantive way. 22 The logic of requiring more than one year of service is that soldiers cannot be expected to rise to a leadership

position where they are given significant responsibility in less than one year.

28

One interesting result is that compulsory military service is positively related to

entrepreneurship, both alone and interacted with r. This is consistent with the view of human

capital acquisition posited by the theory. The military is one institution where very young

people are given significant managerial responsibilities, primarily because the military has

very few older individuals. As a consequence, being in the military is like working in a firm

that has few older workers. The interaction with r is consistent with military service being

an offset to to living in an older country. Although it tough for young workers to acquire

the skills necessary to start a business in an older country, one way that this is accomplished

is by having the young in positions of responsibility in the military.

Finally, column 7, 8, and 9 repeat the analysis of column 1, 2, and 3, respectively, but

include country fixed effects. The precision of the estimates declines, but the coefficients on r

in both subsamples remain important and significant. The fact that coefficients remain

significant is somewhat surprising because most of the variation in r is between countries, not

within countries over time. Apparently, the demographic structure changes sufficiently in

some countries over the period to pick up the demographic effect.

The magnitudes obtained in Table 2 are substantial and suggest an important quantitative

relation of entrepreneurship to demographic structure. Based on the estimates in column 1 of

table 2, a country with r=0.22 (Japan’s value in 2010) would have a predicted

entrepreneurship rate of 0.023. A country with r=-0.10 (the United States’ value in 2010)

would have a predicted entrepreneurship rate of 0.037, which is 61% higher. At the extreme,

a very young country with an r=-2.18 (Uganda’s value in 2010) would have a predicted

entrepreneurship rate of 0.12.

Recall from Corollary 3 that the analysis can be expressed in terms of median age, rather

than r (see empirical implication 5). Using the estimates from column 3 of table 2, a

one-standard deviation decrease in median age (equal to 3.5 years) results in a 2.5 percentage

point increase in the entrepreneurship rate which is over 40 percent of the mean

entrepreneurship rate across countries (equal to .061 in 2010). Based on this result, the

29

difference in the entrepreneurship rate between the youngest (age=31) and oldest (age=44)

country is predicted to be about 9 percentage points or more than twice the mean rate.

5.3. Country-Age Relationships

The cross-country relation of entrepreneurship to the demographic structure is an useful

starting point, but the essence of the theory is best understood at a finer level of detail. Most

of the theoretical implications laid out earlier describe the pattern of entrepreneurship as it

relates to age within a country or to the age-specific variations in the entrepreneurship rates

across countries. In this section, those predictions are tested empirically.

The country-year-age cell is now the unit of observation. The entrepreneurship rate is a

continuous variable that describes the rate of entrepreneurship within a country in a given

survey year for a particular age group. There are 17,554 observations with a mean

entrepreneurship rate of 0.058 and a standard deviation of 0.074 in 2010. About three-fourths

of the cells have positive values for the entrepreneurship rate, with the bulk of the mass

below a 10% entrepreneurship rate.

As noted earlier, the inverted u-shape relation of entrepreneurship to age is observed in

Figure 1. It is also apparent that the entire profile shifts down with the median age of the

country, revealing that younger countries have higher rates of entrepreneurship at every age

than do middle countries.23

This graphical representation of the data provides immediate

support for the theory, but it is important to test the propositions and corollaries more

formally, using the country-age data. Table 3 reports the results.

[Table 3 is here]

As in Table 2, all specifications control for the year with year dummies. Additionally, the

measures of demographics, specifically, r, a and sa, are year-specific. All standard errors are

clustered at the country level and the regressions are weighted by the number of observations

in the relevant cells throughout the table.24

23

These age cutoffs are chosen so that each group has 20, 21, and 16 countries, respectively, i.e., approximately

one-third of the sample for each. 24

Clustering at the country-year level reduced the standard errors.

30

The first column in Table 3 provides a weak test of Proposition 1, which states that

holding age constant, older countries, as reflected by higher levels of r, should have lower

rates of entrepreneurship. The dependent variable is the entrepreneurship rate for the

country-year-age cell. The two independent variables are age, represented as a variable that

goes from zero to one, and r as estimated above.25

Holding age constant, column 1 reveals

that countries that have older populations have lower rates of entrepreneurship. Column 2

does the same analysis, but for OECD countries only. The same pattern is observed. The

prediction of proposition 1 is stronger than that revealed by columns 1 and 2 because it states

that at any given age, countries with higher values of r should have lower rates of

entrepreneurship. The complete interaction structure that is required to test proposition 1 in

its purest form is presented later in Table 4.

Corollaries 1 and 2 of Proposition 1 are addressed by column 3 of Table 3. Corollary 1 is

another version of Proposition 1 with two differences. First, sa replaces r as a measure of the

demographic structure. Second, r varies only across countries but does not vary differently

for the different age groups. There is only one r per country. But sa can vary across countries

for one age group differently than it varies across countries for another age group. There are

45 sa values per country in a given year.

The results reported in column 3 show a positive coefficient on sa, which implies that

given age, the larger the share of the population younger than the age group in question, the

higher is the entrepreneurship rate.26

This is a direct implication of corollary 1. Holding a

constant, the smaller the proportion of the country that is older than the age group in question,

the higher the rate of entrepreneurship. This is consistent with the rank effect. In younger

countries, younger individuals get more opportunity to acquire the human capital necessary to

start a business. As a result, in steady-state, more entrepreneurial human capital is available

in younger countries at any given age.

The coefficient on a is negative, as predicted by corollary 2. For any given structure of sa,

age enters negatively. This is the creativity effect that is the focus of corollary 2. Were it not

25

As before, r is country-year-specific. 26

This is true despite the fact that in the entire sample, a and sa have a correlation over .9.

31

for the necessity of acquiring business skills, younger individuals would have higher rates of

entrepreneurship than older individuals.27

Column 4 repeats the analysis for OECD countries

only. The results are qualitatively the same. Corollaries 1 and 2 are both supported within

OECD-only countries. More complete tests of corollaries 1 and 2 are reported later using in

Table 5, which allows each age group to have its own coefficient on sa, and allows each age to

have its own coefficient as well.

Columns 5 and 6 repeat the analysis of columns 3 and 4, but include country fixed effects.

The inclusion of fixed effects over-controls for country differences because the demographic

structure, insofar as it affects all ages similarly, is picked up by the fixed effect. This is

inappropriate when testing whether demographic structure matters. The model of

demography, as formally stated, is a one parameter, r, model. If one parameter truly captured

all demographic variation among countries and entered in the way proscribed by the

demographic and entrepreneurship model, then fixed effect estimation would disguise all of

the variation that results from demographics (except to the extent that r differs within a given

country over time and here, only for countries with enough multi-year data). To the extent

that a single parameter does a reasonable job at describing the demographic structure, the

effect of structure as measured through sa is suppressed. In fact, the coefficients on sa in

columns 5 and 6 are smaller than those without fixed effects, but maintain the expected signs

and remain important.

Column 7 examines the average (across all countries) effect of age on entrepreneurship

and allows for non-linearities. The inverted u-shape shown in Figure 1 is revealed by the

estimates in column 7, with age entering positively and age-squared entering negatively.

5.4. Age-Specific Estimation

Because there are 45 different age groups per country per year, there are actually 45 tests

that can be run. Specifically, for each age group, the relation of entrepreneurship to either r or

27

This could also reflect time horizon effects, i.e., a longer period over which to reap returns when young, but

the differences in expected present value is unlikely to vary much for young individuals because the

length-of-life effect cuts many years out, when the present value of the differences are small.

32

sa can be estimated. Also, because the country-year-age-specific entrepreneurship rate at one

age is potentially independent of the year-country-age-specific entrepreneurship rate in the

same country-year at another age, there are 45 tests of the theory that can be performed. This

is done in Tables 4 and 5.

In Table 4, each age group is allowed to have its own coefficient on r, which means that

each row in the table provides an estimate of the effect of variations in the age structure, as

reflected in r, on the age-specific entrepreneurship rate. Age 20 is the omitted age group.

[Table 4 is here]

There are three columns. In the first column, only r effects enter. Within each age group,

r varies across countries.

In column 2, r is omitted altogether and only age-dummies enter. This captures the effect

of age, per se, but ignores the variation in demographic structure across countries. As such, it

combines creativity effect and the rank effect to the extent that they are correlated with age.

The rising and then declining age coefficients are another reflection of the inverted-u shape

pattern of entrepreneurship to age that was displayed in Figure 1. Column 2 does not exploit

cross-country variation because age varies in the same way in all countries and demographic

structure is ignored. Consequently, the column 2 estimates merely pick up the average effect

of age on entrepreneurship across all countries. The age dummies display the predicted

inverted-u-shape pattern. (The omitted age is 20.) The highest levels of entrepreneurship are

experienced in the late-twenties through the mid-thirties.

Column 3 includes both age dummies and the age-specific r effects. There are strong

negative effects of r , as predicted by corollary 1. Most important is that the effect of r on the

age-specific entrepreneurship rate is negative and significant in virtually all cases. Each of

these coefficients is based on independent cross-country variation and provides strong

evidence that the demographic structure is closely related to the entrepreneurship rates.

Within every age group, the effect of having an older population is to lower the

entrepreneurship rate for that age group.

33

Table 4 provides strong support for Proposition 1 that the within-age-group rates of

entrepreneurship should vary inversely with r. The regressions are run where the observation

is at the country-year-age level and the dependent variable is the entrepreneurship rate for

that cell. The estimates in column 1 show that countries with older (and declining)

populations have lower rates of new business formation at every age level, that is, for every

row of the table, as the theory predicts. This prediction comes directly from the rank, or

business skills effect and is not related to the simple idea that younger people are more

creative. Younger people in older societies form fewer businesses than younger people in

younger societies. The same holds true at virtually every age level, although it is less

important for older groups than for younger groups.

Column 3 estimates the effect of age and of demographics by including age dummies and

country-specific values of r . Columns 1 and column 3 of table 4 support the prediction of

Proposition 4 that demographic structure matters more near the peak (middle-aged groups)

than it does at either end. The absolute values of the negative coefficients on r first rise and

then decline with age. This is shown in figure 4. Additionally, the creativity effect finds

some weak support in the fact that age enters negatively for the oldest ages in column 3

(although there are a couple of negative coefficients for the very young as well).

[Figure 4 is here]

As is clear, the age-dummies alone do not explain much of the variation, whereas the

age-specific r effects capture most of it. (Compare r-square values across columns.)

Specifically, in column 1, there is a single constant term and the estimates of r are based on

cross country variations in the age composition, not on the within country variation in age.

As a result, the r-squared in column 1 picks up the effect only of cross-country demographic

variation that corresponds to the hypothesized differential business skill acquisition across

countries. It is not a result of the within-country variations in entrepreneurship rates at

different ages. That effect is captured by the regression in column 2, which has a much lower

r-squared (0.09 versus 0.23 in column 1). Finally, the r-squared is higher in column 3 than in

column 1, as it must be, but not by much. Including both age dummies and the age-specific r

34

effects only improves r-squared to 0.25 from the 0.23 of column 1. This is an important result.

It implies that cross-countries differences in entrepreneurship rates observed in table 2 is not

so much a reflection of the young being more creative and entrepreneurial, but of the fact that

the every age group is less entrepreneurial in countries that have an old populations.

It is also possible to allow sa and a to enter directly, which is done in Table 5. These

results provide more evidence on the corollaries to Proposition 1. Variations in sa can only

result from variations in the demographic structure across countries because sa is defined

specific to each age.28

The results of columns 1, 2 and 3 of Table 5 parallel those of columns

1, 2 and 3, respectively, of Table 4. As before, the observation is the county-year-age cell and

the dependent variable is the entrepreneurship rate for that group.

[Table 5 is here]

Column 1 of Table 5 supports corollary 1, just as column 1 of Table 4 supports

proposition 1. The value of sa in each age row is the sa that relates to that age in each of the

countries. Countries that have a higher proportion of individuals younger than the group in

question have higher rates of entrepreneurship.

These results are compelling support of the rank effect, given that each sa coefficient is a

potentially independent test of that effect. The estimates in column 1 are based on different

age-structures in a population. Age itself plays no direct role because the analysis is done

within a given age group for every age.

Similarly, column 3 of Table 5 provides support for corollary 2. The age-dummies tend

to become more negative (relative to age 20) as age grows. Holding sa constant, the younger

the group, the higher the rate of entrepreneurship. This reflects the pure creativity effect

because population structure is held constant by the inclusion of sa.

Note that as in Table 4, the r-squared for the sa only regression is 0.18. Allowing only

within-country age effects to matter yields an r-squared of 0.09. When both effects are

permitted, the r-squared rises to 0.24. Also noteworthy is that using r, as done in Table 4, is

better at explaining variations in entrepreneurship than is using sa directly, as is done in Table

28

Some of the variation in sa might reflect variations over time in sa for a given country when there are

multiple years surveyed for that particular country.

35

5. Recall that r is an estimated variable, which might be expected to be noisier than the sa on

which r is based. However, the estimated r uses the information across all age groups within

a country. As a consequence, r may be a better indicator of the demographic structure

relevant for a given age group than that age group’s own sa.

5.5. Entrepreneurship Rates Peak in Middle Age

Figure 1 showed that a country’s entrepreneurship rates rises and then falls with age, as

predicted by Proposition 3. The implication also borne out by column 7 of Table 3, where a

enters positively and a2 enters negatively resulting in an inverted-U which peaks at age 32.

29

Column 7 captures the relation of entrepreneurship to age for the average country, but it

is also possible to use the estimates to simulate the age-entrepreneurship profile that would

exist for any hypothetical country with given age parameter r.

The analytic form of equation (10) can be used obtain a predicted value of as for any

hypothetical value of r. This value, coupled with the estimates of Table 3, column 3, can be

used to derive a predicted entrepreneurship rate at each age, a. Then it is straightforward to

plot the estimated entrepreneurship rates by age for any country with a hypothetical value of

r . This is done for a few values of r and the results are displayed in figure 5.

[Figure 5 is here]

For r = -.5, the rate peaks at age 30 and has the predicted inverted u-shape that shows up

in the actual data. This is neither mechanical nor a direct consequence of including non-linear

terms. Non-linear terms in a are excluded from the direct calculation. The inverted

u-shape results because the creativity effect, captured by a, initially outweighs the rank effect,

captured by sa . That turns around at some age near 30.

The u-shape depends on the value of r and becomes more pronounced as r falls.

Compare the graphs in figure 5 for r=-2.19 (Uganda) to that for r=-1 and then to that for r=-.5.

the entrepreneurship rates are higher and display a more pronounced peak the lower is r, as

predicted by the theory. The inverted u-shape disappears for higher values of r. For

29

20+0.045/(2*0.085)*(64-20) = 32

36

example, the graph for Japan, with r=.22 displays entrepreneurship monotonically decreasing

in age. (The actual rates for Japan, however, do display a slight inverted U-shape, as

predicted by the theory.)

Finally, Proposition 4 states that middle-age entrepreneurship rates are most sensitive to

changes in demographic structure across countries, as reflected in r. As an empirical

proposition, this holds, again using the estimates from column 3 of Table 3. Because

ar

F

ar

s

s

E

ar

Eas

a

a

222

and because βsa is positive in column 3 of Table 3, it follows, using lemma 2, that the

entrepreneurship rate among the middle-age shift down more with increasing r than at older

and younger ages. This is also what is observed in the raw data, as was shown earlier in

Figure 1.

5.6. Robustness Checks

Recall that entrepreneurship defined an entrepreneurial event as owning a business that is

less than 42 months old, which at the time of the survey, pays wages. This section speaks to

the sensitivity of the results to the definitions of entrepreneurship chosen.

The results are reported in Table 6. Panel A of Table 6 replicates the regressions of Table

2, columns 5 and 6, for three definitions of entrepreneurship. The specifications of columns 5

and 6, Table 2, are the most general. They included all controls at the country level. Only

the coefficients of interest are reported. The alternative definitions include: (1) early-stage

entrepreneur plus those businesses that have died in the last twelve months labeled

“esentr_shd”, (2) nascent businesses that are less than 42 months old but do not pay wages

labeled “nascent”, (3) nascent plus early-stage entrepreneurs labeled “ttl_esentr.”

Altering the definition of entrepreneurship has no substantive effect on the conclusions.

The coefficient on r remains negative and significant by all definitions, although the

magnitudes change somewhat. (Of course, the dependent variables also have different means

and variances, which would imply different coefficients on the independent variables even if

37

the effects were identical.) As before, the higher the population shrinkage parameter, the

lower the overall entrepreneurship rate..

Panel B of Table 6 repeats the analysis of Table 3, columns 3 and 4, using the three

alternative definitions of entrepreneurship. As with panel A, the findings of Table 3 are

robust to alternative definitions of entrepreneurship. The coefficients change somewhat, but

the qualitative results remain unaltered.

Special attention is given to high aspirational entrepreneurship using various criteria to

define high aspirations. These include entrepreneurial events where the founder plans to

hire more than five or more than ten employees during the next ten years, which distinguishes

these entrepreneurial events from self-employment where an individual works alone or with a

very small number of others.30

Additionally, individuals who plan to produce new products

or services are singled out as entrepreneurs.

The results are reported in Table 7, which parallels the structure of Table 6. Panel A of

Table 7 does the cross country comparison and is a robustness check on Table 2, whereas

panel B of Table 7 is based on the country- age cells and is a robustness check on Table 3.

Once again, the coefficients on r remain negative and significant in the country level

regressions (panel A), and the coefficients on sa remain positive and significant in

country-age level regressions (panel B). These findings suggest that the results hold for

entrepreneurs with high aspirations as well.

5.8 Summary of Predictions and Findings

The theory yielded specific empirical relations that should be observed in the data.

Here, the predictions and findings are summarized.

1. Proposition 1 predicts that at any given age, entrepreneurship decreases as the

population ages, as measured by the shrinkage parameter r. Table 3 and especially

Table 4 provide strong support. In Table 3, r has the predicted negative effect on

entrepreneurship, holding age constant. In Table 4, for column 1, at every age,

30

Recall, however, that the definition of entrepreneurship used throughout the paper should exclude sole

self-employed individuals, because to qualify as an entrepreneur the firm in question must be paying wages.

38

entrepreneurship rates are negatively related to r. The same is almost always true in

Table 4, column 3, where each age is allowed to have its own intercept.

2. Corollary 1 states that entrepreneurship should rise in sa, given age. This is borne out

by Table 3, columns 3, 4, 5 and 6, and by Table 5, which, analogous to Table 4,

shows that entrepreneurship at any given age is negatively related to sa.

3. Corollary 2 says that entrepreneurship should be inversely related to age, given sa.

Table 3, columns 3, 4, 5, and 6, support this claim, as does Table 5, which shows age

coefficients in column 2 tending to decline with age.

4. Proposition 2 predicts that the overall entrepreneurship rate declines as the population

ages, as measured by r. Table 2, columns 1, 2, 7 and 8 support this prediction.

5. Corollary 3 is a restatement of proposition 2, but cast in terms of median age instead

of r. Table 2, columns 3 and 4 bear out this prediction.

6. Proposition 3 predicts an inverted u-shaped relation of entrepreneurship to age.

Column 7 of Table 2 provides direct support for this proposition as does the raw data

shown in Figure 1 and the estimated age-entrepreneurship profiles in Figure 5.

7. Proposition 4 predicts that entrepreneurship rates decline most for middle-aged

workers as a population ages. This is borne out by Figures 1 and 5 and by the

analytic result that based on lemma 2 coupled with the positive coefficient on βsa in

column 3 of Table 3.

5.8 Identification

The approach used is one that assumes the demographic structure is given and

pre-determined. There are two potential concerns. First, one might worry that causation

might run in the other direction, where entrepreneurship causes the demographic structure,

rather than the demographic structure causing entrepreneurship. Second, there may be other

factors that are correlated with demographic structure that might generate a relation of

entrepreneurship to demographic structure other than the one featured in this analysis.

Before addressing identification at a more technical level, it is important to point out that

the demographic theory provides very specific implications (at least seven) that have been

39

tested and found to hold. Some of these, and especially the collection of them, are not easily

generated by other theories. The fact that seven-out-of-seven of the empirical implications are

supported strongly by the estimates, irrespective of the country sample and functional form,

is helpful, but there are additional reasons why the evidence points toward this theory or

some close variant of it.

First, the relation of entrepreneurship to demographics almost certainly reflect the latter

causing the former, not the former causing the latter. The demographic structure is

determined well before a birth cohort grows to an age sufficient to be entrepreneurs. Except

for immigration, the proportion of the population that is over any given age is an exogenous

and pre-determined variable from the point of view of the entrepreneurship rate. It would be

difficult, if not impossible, to come up with a story that had an exogenously given

entrepreneurship rate in 2009 determining the number of births that occurred in a country in

1979. Even the current entrepreneurship rate is unlikely to have any strong effect on current

birth and mortality rates.

It is possible, however, that a country that is conducive to entrepreneurship could attract

a large number of immigrants, particularly young ones. This effect is likely to be small for

two reasons. First, in most countries, the age of immigrants is not so different from the age of

the population as a whole. For example, the Department of Homeland Security reports that in

the United States, the median age of Green Card holders is five years lower than that of the

native born population. Green Card holders are those who have immigrated to the US, but

have not obtained citizenship status.31

Although five years is significant, Green Card holders

account for only about one-thirtieth of all those living in the US, which means that the five

year difference would lower the median by about 1/6 of a year. Indeed, among the 2010

sampled countries, median ages (among population aging between 20 to 64) range from 31 to

44 years old, with a standard deviation of almost four years. As a consequence, immigration,

even on a large scale, is unlikely to cause much bias in the estimates.

31

Lazear (2013) and Office of Immigration Statistics (2012).

40

The second point supporting the theory postulated here is that it is difficult to find

alternative explanations for the fact that sa has the predicted positive effect on

entrepreneurship, r has the predicted negative effect, and that these effects are pervasive. The

next section considers some possible alternative theories.

Alternative Mechanisms

The most novel and compelling finding is that within every age group, older countries

have lower rates of entrepreneurship. The interpretation offered in this study is that older

workforces reduce the ability of young workers to acquire business skills, but no direct

evidence on this interpretation has been offered. There may be other reasons why older

countries would have lower rates of entrepreneurship (at all ages) that differ from the one

suggested here. A few possibilities come to mind.

An alternative is that countries with older populations are further along in the

development cycle and have lower returns to starting new businesses. Although it is not clear

that the premise even holds, it is possible to provide evidence that speaks to this hypothesis.

Columns 5 and 6 of Table 2 allow other factors to affect the country-wide entrepreneurship

rate. Per-capita GDP would seem to be a good proxy for stage of development. It does not

have a quantitatively meaningful effect on the results and the coefficient on r remains

negative, important, and statistically precise.

A second and related possibility is that even if the country is not further along in the

development cycle, the age distribution may reflect the growth potential of countries.

Specifically, older countries may not be good places to start new businesses because future

growth prospects are poorer than they are in younger countries. If this were the case, then it is

possible that at every age, entrepreneurship rates would be negatively related to the median

age or to r. The country fixed-effect regressions address this in part because the country

fixed effect would tend to pick up the growth prospects. But it is still possible that

unanticipated changes in the demographic structure could affect the incentives to start

businesses.

41

Two findings argue against this view. First, columns 5 and 6 of Table 2 include the

five-year average GDP growth rate, which one would expect would be a reflection of

investment potential. The growth rate does not do much to explain entrepreneurship (and is

not significant in the full sample), nor does it eliminate or substantially reduce the importance

of r in explaining entrepreneurship. Second, some of the implications of the skill

acquisition story are not obvious implications of the view that the age structure of a country

proxies growth potential. In particular, the inverted u-shape relation of entrepreneurship to

age comes about because two effects work in opposite directions. The decline in creativity

with age is at first swamped by the entrepreneurship skills that are acquired with work

experience, which results in entrepreneurship rates that increase at first in age. When skill

acquisition tapers off enough, the declining creativity effect swamps the skill acquisition

effect, resulting in declining entrepreneurship rates with age. It is the combination of the

two effects that yields the prediction of an inverted u, borne out by the results. There is no

obvious reason why an inverted u-shape pattern would be produced in an environment where

investment potential was proxied by the age distribution. Similarly, the prediction that r

affects middle-aged entrepreneurship rates by more than the old or young is not an obvious

prediction of the age-distribution-proxying investment potential story, although ad hoc

explanations could surely be offered.

5.9 Extensions

Industries

The analysis here is aggregate in the sense that the country’s demographic structure is

assumed to be relevant for all industries in the country. That assumption is made by

necessity because data are not readily available across countries that would relate to

industry-specific demographics. Such data could prove relevant and provide interesting

extensions of the analysis.

For example, the age structure of the United States might not be a good proxy for the

age-structure of the technology industry during the dot.com boom. Because technical skills

42

in computing were needed and because those skills tended to be concentrated among the

young, the median age in the tech industry was lower than that for the working population as

a whole. As a consequence, young workers in technology held positions with more

responsibility than would have been predicted given their rank in the overall age distribution.

From the point of view of bias, this is not likely to be a problem, at least with respect to

table 2. Because this relates to a mismeasured independent variable (r or median age), the

coefficients reported in that table are biased toward zero. The estimated effects, if anything,

understate the true effects.

The more important issue is that more direct tests of the theory are available, at least

conceptually. As a result, a useful extension would be to do industry-based analysis, using

more appropriately defined demographic variables. The rank and creativity effects can be

separated from overall industry effects by doing within-age across industry analysis.32

Individuals

It is also possible to use individual-based data to test the hypotheses of the model. A

particular individual’s experience should be a predictor of entrepreneurship. In Lazear

(2004, 2005), having broad prior experience was used as a predictor of entrepreneurial

activity. But additionally, the theory of the current analysis suggests that those who had

positions that were high up in a firm’s hierarchy should be more likely to be entrepreneurs

later on.

This prediction can be tested using the Stanford MBA data described in depth in Lazear

(2005). Those data contain information on the number of others supervised in prior

positions. A simple check reveals that those who supervised more workers on their prior

jobs are more likely to have founded a business on subsequent job events. This is consistent

with the theory here, but is not conclusive for the obvious reason that those who were

assigned to supervisory roles may have already possessed the talents necessary to later found

32 For example, entrepreneurship is high in technology and technology is a young persons industry for the the same reason:

technology is a new field. But the difference between countries in the entrepreneurship rate at any given age, based on that

country’s age structrure allows the newness effect to be separated from the rank and creativity effects.

43

businesses. An appropriate test would examine the subsequent entrepreneurial experiences

of people who were randomly thrust into supervisory positions.

6. Conclusion

Using a human capital framework like that proposed by Becker, coupled with an

economic focus on fertility patterns, again studied by Becker, a connection between

demographics and entrepreneurship is proposed. To become an entrepreneur, an individual

needs to have an idea for a business, which may be more prevalent among the young. But

making a new business a success also requires skills that are best acquired through on-the-job

training. In an aging society, the important positions in firms are likely to be dominated by

older individuals, whereas in a younger society, young workers may get the opportunity to

experience situations that will better enable them to start their own businesses. As a

consequence, a young society allows more opportunity for workers to acquire business skills

early in their careers, which implies that they will be more skilled at any given age.

The theory proposes a basic demographic equation that relates the age structure of a

society to one parameter, namely the population shrinkage (or conversely, growth) rate. The

one parameter structure fits the demographic data very well, explaining almost perfectly the

demographic structure across a large number of countries. More important, the theory that

is based on this structure has a number of testable implications that apply to entrepreneurship,

all of which are borne out using an unusual dataset called the Global Entrepreneurship

Monitor.

Most basic of the predictions is that the older the population, the lower is the overall rate

of entrepreneurship. The model predicts this holds both with respect to the population

shrinkage parameter, which measures the speed with which a population ages, as well as the

median age of a country. Both relationships show up clearly in the data and are robust to

various definitions of entrepreneurship. They are not eliminated, or even substantially

weakened, by the inclusion of other variables that might relate to alternative theories linking

entrepreneurship to age structure of the population. In particular, the hypothesis that the age

structure is a mere proxy for stage of development or growth potential is rejected.

44

The model implies that entrepreneurship should display an inverted u-shape relation to

age. The very young do not possess the business skills necessary to start a business and the

very old lack the creativity or energy to engage in entrepreneurship. The inverted u-shape

holds throughout and is found in almost all countries sampled.

Perhaps the prediction most specific to this model is that at any given age, countries that

have younger populations have higher entrepreneurship rates among that age group than do

countries with older populations. This is because of the “rank” effect, which implies that

younger countries allow for higher rates of human capital acquisition than older countries.

As a consequence, workers in younger countries have more skills relevant for

entrepreneurship at every age. This is found to hold strongly and the finding is robust to all

empirical specifications. Furthermore, the rank effect accounts for the bulk of the difference

in entrepreneurship rates across countries. The age composition effect, which says that

younger countries have higher rates of entrepreneurship because they have more younger

workers, is much less important in explaining cross country variation in entrepreneurship

rates.

Finally, the entrepreneurship rates of the middle-aged are predicted by the framework to

be the most sensitive to the rate at which a population is aging. Entrepreneurship rates of

the very young and very old are predicted to be more similar across countries than are the

entrepreneurship rates of the middle aged. The theory implies that middle-aged

entrepreneurship rates should be most sensitive to variations in the demographic structure.

This result is found to hold.

The linkage between entrepreneurship and demographics that is studied here goes back

to Becker’s early work on both human capital and fertility. More than fifty years after the

work was done, new implications of his profound theories continue to be discovered and

borne out.

45

References

Acemoglu, Daron, Ufuk Akcigit, and Murat Alp Celik. “Young, Restless and Creative:

Openness to Disruption and Creative Innovations,” NBER Working Paper, 19894, 2014.

Acht. J., J. Stam., A.R. Thurik., and I. Verheul. 2004. “Business Ownership and

Unemployment in Japan.” Discussion Papers on Entrepreneurship, Growth and Public

Policy No. 0904. Max Planck Institute for Research into Economic Systems. Jena,

Germany.

Barro, Robert and Jong-Wha Lee. 2010. “A New Data Set of Educational Attainment in the

World. 1950-2010.” Journal of Development Economics 104(C): 184-198

Becker, Gary S. 1960. “An Economic Analysis of Fertility.” In Demographic and Economic

Change in Developed Countries, eds. George B. Roberts, Universities-National

Bureau Committee for Economic Research. Chapter 7, pp. 209-240. Columbia

University Press: New York, NY.

Becker, Gary S. 1962. “Investment in Human Capital: A Theoretical Analysis.” Journal of

Political Economy 70(5): 9-49.

Becker, Gary S. 1975. “Human Capital: A Theoretical and Empirical Analysis, with Special

Reference to Education.” 2nd edition. Columbia University Press: New York, NY.

Blanchflower, David G., and Bruce D. Meyer. 1994. “ A Longitudinal Analysis of the Young

Self-Employed in Australia and the United States.” Small Business Economics 6(1):

1-19.

Blanchflower, David G., and Andrew J. Oswald. 2009. “What Makes a Young

Entrepreneur? ” in International Handbook on Youth and Young Adulthood, ed. Andy

Furlong. Routledge: London, UK.

Bloom, David E., David Canning, and Gunther Fink. 2010. "Implications of Population

Ageing for Economic Growth." Oxford Review of Economic Policy 26(4): 583-612.

Bosma, Niels, Alicia Coduras, Yana Litovsky, and Jeff Seaman. 2012. “Global

Entrepreneurship Monitor Manual.” Version 2012-9: May. Retrieved From:

http://www.gemconsortium.org/docs/download/2375

Dore, R. 1996. "The End of Jobs for Life?: Corporate Employment Systems: Japan and

Elsewhere." Centre for Economic Performance Occasional Paper No. CEPOP11.

London School of Economics and Political Science. London, UK.

Florida, R. 2002. The Rise of the Creative Class: And How It’s Transforming Work, Leisure,

Community and Everyday Life. Basic Books: New York, NY.

46

Galenson, David W. 2001. Painting Ouside the Lines: Patterns of Creativity in Modern Art.

Havard University Press: Cambridge, MA.

Global Entrepreneurship Research Association (GERA). 2010. Global Entrepreneurship

Monitor (2001-2010). Adult Population Survey. Retrieved From:

www.gemconsortium.org.

Granovetter, Mark. 1985. "Economic Action and Social Structure: The Problem of

Embeddedness." The American Journal of Sociology 91(3): 481-510

Greve, Arent and Janet W. Salaff. 2003. "Social Networks and Entrepreneurship."

Entrepreneurship, Theory and Practice 28(1): 1-22.

Herrmann, Michael. 2012. “Population Aging and Economic Development: Anxieties and

Policy Responses.” Journal of Population Aging 5(1): 23-46.

Heston, Alan, Robert Summers, and Bettina Aten, Penn World Table Version 7.1, Center for

International Comparisons of Production, Income and Prices at the University of

Pennsylvania, Nov 2012.

Karlin, Andrew R. 2013. “The Entrepreneurship Vacuum in Japan: Why It Matters and How

to Address It.” Knowledge at Wharton. Retrieved From:

http://knowledge.wharton.upenn.edu/article.cfm? articleid=3145.

Kaufman, A. S. and J. L. Horn. 1996. “Age Changes on Tests of Fluid and Crystallized

Ability for Women and Men on the Kaufman Adolescent and Adult Intelligence Test

(KAIT) at Ages 17-94 Years.” Archives of Clinical Neuropsychology 11(2): 97-121.

Lafontaine, Francine and Kathryn Shaw. 2014. “Serial Entrepreneurship: Learning by

Doing?” NBER Working Paper No. 201312. National Bureau of Economic Research.

Cambridge, M.A.

Lotka, Alfred J. 1922. "The Stability of the Normal Age Distribution." Proceedings of the

National Academy of Sciences of the United States of America 8(11): 339-345.

Lazear, Edward P. 2004a. "Balanced Skills and Entrepreneurship." American Economic

Review 94(2): 208-211.

Lazear, Edward P. 2005. “Entrepreneurship,” Journal of Labor Economics, 23:4

(October 2005): 649-80.

Lazear, Edward P. 2004b. “The Peter Principle: A Theory of Decline.” Journal of Political

Economy 112(S1): S141-S163.

Lazear, Edward P. 2013. “The Young, the Restless and Economic Growth.” Wall Street

Journal. Retrieved From:

47

http://online.wsj.com/news/articles/SB100014240527023032936045792539325346010

74.

Mondragon-Velez, Camilo. 2009. “The Probability of Transition to Entrepreneurship

Revisited: Wealth, Education and Age.” Annals of Finance 5(3-4): 421-441.

Office of Immigration Statistics. 2012. Yearbook of Immigration Statistics: Table 8. U.S.

Dept. of Homeland Security. Retrieved From:

http://www.dhs.gov/yearbook-immigration-statistics.

Parker, C. Simon. 2006. The Economics of Self-Employment and Entrepreneurship.

Cambridge University Press: New York, NY.

Prettner, Klaus. 2013. “Population Aging and Endogenous Economic Growth.” Journal of

Population Economics 26(2): 811-834.

Ruth, K.-E. and J. E. Birren. 1985. “Creativity in Adulthood and Old Age: Relations to

Intelligence. Sex and Mode of Testing.” International Journal of Behavioral

Development 8(1): 99-109.

Ryan. J. J., J. M. Sattler and S. J. Lopez. 2000. “Age Effects on Wechsler Adult Intelligence

Scale-III Subsets.” Archives of Clinical Neuropsychology 15(4): 311-317.

Schwarz, A. and Asli Demirgue-Kunt. 1999. “Taking Stock of Pension Reforms around the

World.” Social Protection Discussion Paper Series No. SP 9917. The World Bank:

Washington, D.C.

Shire, Karen. 2008. “Gender Dimensions of the Aging Workforce." In The Demographic

Challenge: A Handbook about Japan, eds. Florian Coulmas, et. al. Chapter 53, pp.

963-962. Koninklijke Brill NV: Leiden, Netherlands.

Strokova, Victoria. 2010. “International Property Rights Index: 2010 Report.” Property

Rights Alliance. The Liberal Institute of the Friedrich Naumann Foundation:

Potsdam-Babelsberg.

Vogel, Edgar, Alexander Ludwig, and Axel Borsch-Supan. 2013. “Aging and Pension

Reform: Extending the Retirement Age and Human Capital Formation.” NBER

Working Paper No. 18856. National Bureau of Economic Research. Cambridge, MA.

Wang, J.-J. and A. S. Kaufman. 1993. “Changes in Fluid and Crystallized Intelligence across

the 20- to 90-Year Age Range on the K-BIT.” Journal of Psychoeducational

Assessment 11(1): 29-37.

Wellman, B. (Ed.). 1999. Networks in the Global Village: Life in Contemporary Communities.

Westview Press: Boulder, CO.

48

Figure 1. Countries with young and old labor forces

0

.02

.04

.06

.08

.1.1

2.1

4

En

trep

ren

eurs

hip

Rat

e

20 30 40 50 60Age

Young countries (Median age<=37)

Middle countries (37<Median age<=41)

Old countries (Median age>41)

Note: This figure presents the raw relationship entrepreneurship rates and age. Using the Global

Entrepreneurship Monitor 2010 data, we group all countries into three categories: (1) young countries,

defined as having a median age younger than 38, (2) middle countries, defined as having a median age

between 38 and 41, and (3) old countries, defined as having a median age greater than 41. These age

cutoffs are chosen so that each group has 20, 21, and 16 countries, respectively, i.e., approximately

one-third of the sample for each.

49

Figure 2. Age Distribution F(a,r)

0 0.2 0.4 0.6 0.8 10

0.5

1

1.5

2

2.5

Normalized age (0=age 20, 1=age 65)

r = -2.18(Uganda)

r = 0

r = 0.22(Japan)

0 0.2 0.4 0.6 0.8 1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Normalized age (0=age 20, 1=age 65)

Pop

ula

tio

n P

rop

ort

ion

r = -2.18(Uganda)

r = 0

r = 0.22(Japan)

(a) p.d.f (b) c.d.f

50

Figure 3. Entrepreneurship Rate and Demographics (2010)

AGO

ARG

AUS

BEL

BIH

BOL

BRA

CHE

CHL

CHNCOL

CRI

DEUDNK

ECU

EGY

ESP

FINFRAGBR

GHA

GRC

GTM

HRVHUNIRL

IRN

ISLISR

ITA

JAM

JPN

KORLVA

MEXMKDMNEMYS

NLDNOR

PAK

PER

PRT

ROURUS

SAU

SVNSWE

TTO

TUNTUR TWN

UGA

URY

USA

VUT

ZMB

0

.05

.1.1

5.2

.25

Entr

ep

reneu

rship

rate

for

entire

co

untr

y

30 35 40 45Median age (20-64)

AGO

ARG

AUS

BEL

BIH

BOL

BRA

CHE

CHL

CHNCOL

CRI

DEUDNK

ECU

EGY

ESP

FINFRA

GBR

GHA

GRC

GTM

HRVHUNIRL

IRN

ISLISR

ITA

JAM

JPN

KORLVA

MEXMKDMNEMYS

NLDNOR

PAK

PER

PRT

ROURUS

SAU

SVNSWE

TTO

TUNTUR TWN

UGA

URY

USA

VUT

ZMB

0

.05

.1.1

5.2

.25

Entr

ep

reneu

rship

rate

for

entire

co

untr

y

-2 -1.5 -1 -.5 0Parameter r

(a) Overall sample (b) Overall sample

AUS

BEL

CHE

CHL

DEUDNK

ESP

FIN

FRA

GBRGRC

HUN

IRL

ISL

ISR

ITA

JPN

KOR

MEX

NLD

NOR

PRT

SVNSWE

TUR

USA

0

.02

.04

.06

.08

Entr

epre

neu

rship

rat

e fo

r en

tire

countr

y

35 37 39 41 43 45Median age (20-64)

AUS

BEL

CHE

CHL

DEUDNK

ESP

FIN

FRA

GBRGRC

HUN

IRL

ISL

ISR

ITA

JPN

KOR

MEX

NLD

NOR

PRT

SVNSWE

TUR

USA

0

.02

.04

.06

.08

Entr

epre

neu

rship

rat

e fo

r en

tire

countr

y

-1.2 -.9 -.6 -.3 0 .3 .6Parameter r

(c) OECD sample (d) OECD sample

51

Figure 4. Absolute value of the coefficients on r by Age in the Age-specific Entrepreneurship Rate

Regression

0.0

00.0

20.0

40.0

6

Abso

lute

valu

e o

f co

effic

ient

on r

20 30 40 50 60Age

Note: This figure plots the magnitude of the age-specific coefficient on r reported in column 1 of

Table 4.

52

Figure 5. Predicted Effect of Demographics, r, on Entrepreneurship and Age

20 25 30 35 40 45 50 55 60 65

0

0.02

0.04

0.06

0.08

0.1

0.12

Age

Pre

dic

ted E

ntr

epre

neurs

hip

Rate

r = -2.18 (Uganda)

r = -1

r = -0.5

r = 0.22 (Japan)

Note: This figure presents the predicted entrepreneurship rate by age for different value of r, based on

the coefficient estimates in column 3 of Table 3 and the predicted value of sa.

53

Table 1. Summary Statistics

Overall (N=57) OECD (N=26) Non-OECD

(N=31)

Mean Std Mean Std Mean Std

Entrepreneurship rate

Alternative definitions

Early-stage, pay wage (esentr)

0.061 0.063

0.030 0.012

0.087 0.075

Early-stage + Shutdown (esentr_shd)

0.106 0.107

0.055 0.019

0.150 0.130

Nascent, not-pay wage (nascent)

0.065 0.065

0.035 0.023

0.090 0.077

Total Early Stage (tt_esentr)

0.121 0.108

0.064 0.029

0.170 0.126

With high aspiration

Plan to hire 5 in 5 years

0.010 0.009

0.006 0.004

0.013 0.011

Plan to hire 10 in 5 years

0.005 0.005

0.003 0.002

0.006 0.006

New product/market

0.011 0.015

0.007 0.006

0.015 0.020

Demographic (among age 20-64)

Cohort shrink rate ( r)

-0.63 0.64

-0.17 0.35

-1.01 0.58

Average age

39.55 2.52

41.34 1.40

38.04 2.25

Median age

38.74 3.50

41.19 1.96

36.68 3.18

Percentage of young (20-45)

0.52 0.09

0.46 0.05

0.57 0.08

Other Characteristics

GDP per capita $ 19032 13230

30036 9819

9802 7306

College enrollment rate % 11.23 6.73

15.59 6.28

7.58 4.64

Start-up cost (% of GNP per capita) % 17.01 34.51

6.47 6.90

25.86 44.80

Property right index

6.02 1.34

7.10 1.00

5.11 0.79

Military service more than a year 0.09 0.29 0.08 0.27 0.10 0.30

Note:

The summary statistics presented in this table are of the countries surveyed in 2010 in GEM. Refer to Table B1 for the list of the

countries.

54

Table 2. Country-Year Level Entrepreneurship Rate Regression

Dep. Var. Entrepreneurship Rate

Sample All OECD All OECD All OECD All OECD All

specifics (1) (2) (3) (4) (5) (6) (7) (8) (9)

r -0.041 -0.023

-0.025 -0.030 -0.047 -0.050

[0.007]*** [0.007]***

[0.009]** [0.008]*** [0.024]* [0.023]**

Median age

-0.007 -0.004

-0.002

(age 20-64)

[0.001]*** [0.001]***

[0.002]

log(GDPpc)

-0.020 -0.006

[0.007]*** [0.009]

Tertiary

0.062 0.032

[0.041] [0.029]

Start-up Cost

0.024 0.016

[0.015] [0.018]

IPRI

0.004 0.008

[0.004] [0.002]***

GDP growth rate

0.002 0.001

(Average of the past 5 years)

[0.001] [0.001]

Military service > 1 year

0.038 0.042

[0.022]* [0.010]***

(Military service > 1 years) x r

0.047 0.064

[0.016]*** [0.032]*

Constant 0.010 0.022 0.318 0.17 0.169 0.016 -0.024 0.027 0.128

[0.004]** [0.004]*** [0.051]*** [0.043]*** [0.063]*** [0.080] [0.040] [0.010]** [0.071]*

Year FE Y Y Y Y Y Y Y Y Y

Country FE Y Y Y

Obs 393 230 393 230 393 230 393 230 393

R-square 0.40 0.24 0.37 0.19 0.53 0.38 0.91 0.72 0.90

Note:

Observations are weighted by the number of individuals who make up each country-year cell.

Standard errors clustered at the country level are in parentheses.

* significant at 10%; ** significant at 5%; *** significant at 1%

55

Table 3. Country-Age Regressions

Dep. Var. Entrepreneurship rate within country-age cell

Sample

specifics All OECD All OECD All OECD All

(1) (2) (3) (4) (5) (6) (7)

a -0.028 -0.023 -0.376 -0.191 -0.189 -0.104 0.045

[0.003]*** [0.003]*** [0.057]*** [0.049]*** [0.025]*** [0.036]*** [0.009]***

r -0.038 -0.021

[0.007]*** [0.007]***

sa

0.336 0.164 0.156 0.079

[0.054]*** [0.048]*** [0.024]*** [0.036]**

a2

-0.085

[0.008]***

Constant 0.025 0.033 0.038 0.04 0.046 0.04 0.035

[0.004]*** [0.004]*** [0.003]*** [0.003]*** [0.003]*** [0.002]*** [0.004]***

Year

dummies Y Y Y Y Y Y Y

Country FE

Y Y

Obs 17554 10309 17554 10309 17554 10309 17554

R-square 0.26 0.12 0.25 0.12 0.56 0.24 0.11

Note:

Observations are weighted by the number of individuals who make up each country-age-year cell.

Standard errors clustered at the country level are in parentheses.

* Significant at 10%; ** significant at 5%; *** significant at 1%.

56

Table 4. Age-specific Coefficient on r

Dep. Var. Entrepreneurship Rate

(1) (2) (3)

Coefficient on r for age

21 -0.031***

-0.039***

22 -0.034***

-0.040***

23 -0.042***

-0.046***

24 -0.048***

-0.049***

25 -0.048***

-0.046***

26 -0.054***

-0.047***

27 -0.054***

-0.049***

28 -0.059***

-0.048***

29 -0.063***

-0.050***

30 -0.055***

-0.046***

31 -0.061***

-0.051***

32 -0.062***

-0.049***

33 -0.058***

-0.044***

34 -0.061***

-0.051***

35 -0.056***

-0.047***

36 -0.056***

-0.049***

37 -0.051***

-0.043***

38 -0.053***

-0.045***

39 -0.049***

-0.042***

40 -0.044***

-0.039***

41 -0.045***

-0.041***

42 -0.045***

-0.039***

43 -0.046***

-0.044***

44 -0.047***

-0.044***

45 -0.047***

-0.044***

46 -0.036***

-0.032***

47 -0.039***

-0.042***

48 -0.033***

-0.035***

49 -0.031***

-0.031***

50 -0.035***

-0.040***

51 -0.039***

-0.046***

52 -0.028***

-0.032***

53 -0.024***

-0.026***

54 -0.030***

-0.036***

55 -0.026***

-0.034***

56 -0.021***

-0.027***

57

57 -0.018***

-0.024***

58 -0.025**

-0.038***

59 -0.026***

-0.041***

60 -0.018**

-0.034***

61 -0.008

-0.021***

62 -0.014**

-0.031***

63 -0.009

-0.025***

64 -0.005

-0.019***

Coefficient on age group dummies

21

0.006*** -0.015***

22

0.010*** -0.012**

23

0.015*** -0.010*

24

0.020*** -0.007

25

0.022*** -0.004

26

0.028*** 0.003

27

0.027*** 0.000

28

0.032*** 0.007

29

0.036*** 0.009

30

0.029*** 0.004

31

0.033*** 0.006

32

0.035*** 0.008

33

0.033*** 0.009**

34

0.033*** 0.005

35

0.030*** 0.005

36

0.029*** 0.002

37

0.027*** 0.003

38

0.028*** 0.003

39

0.025*** 0.003

40

0.022*** 0.001

41

0.020*** -0.002

42

0.023*** 0.002

43

0.020*** -0.004

44

0.022*** -0.001

45

0.021*** -0.003

46

0.016*** -0.001

47

0.014*** -0.009*

48

0.011*** -0.008*

49

0.012*** -0.005

50

0.010*** -0.011**

51

0.012*** -0.013**

52

0.008*** -0.009*

53

0.006** -0.008*

58

54

0.007*** -0.012**

55

0.005** -0.013**

56

0.003 -0.012***

57

0.001 -0.012***

58

0.001 -0.019***

59

0.001 -0.021***

60

-0.005* -0.023***

61

-0.008*** -0.019***

62

-0.007*** -0.024***

63

-0.009*** -0.023***

64

-0.011*** -0.021***

Year FE Y Y Y

Obs 17,554 17,554 17,554

R-square 0.232 0.086 0.248

Note: Standard errors are clustered at the country level, but

not reported to save space. * Significant at 10%; **

significant at 5%; *** significant at 1%.

59

Table 5. Age-specific Coefficient on sa

Dep. Var. Entrepreneurship Rate

(1) (2) (3)

Coefficient on sa for age

21 2.216***

2.783***

22 1.175***

1.449***

23 0.855***

1.132***

24 0.689***

0.928***

25 0.562***

0.717***

26 0.508***

0.608***

27 0.432***

0.559***

28 0.403***

0.491***

29 0.374***

0.476***

30 0.311***

0.406***

31 0.299***

0.415***

32 0.278***

0.384***

33 0.252***

0.328***

34 0.234***

0.368***

35 0.213***

0.336***

36 0.197***

0.347***

37 0.180***

0.304***

38 0.173***

0.316***

39 0.159***

0.292***

40 0.144***

0.273***

41 0.135***

0.291***

42 0.134***

0.277***

43 0.124***

0.327***

44 0.123***

0.326***

45 0.117***

0.333***

46 0.104***

0.252***

47 0.098***

0.347***

48 0.090***

0.296***

49 0.088***

0.291***

50 0.084***

0.354***

51 0.083***

0.425***

52 0.076***

0.293***

53 0.072***

0.278***

54 0.071***

0.391***

55 0.067***

0.406***

56 0.062***

0.385***

57 0.059***

0.329***

60

58 0.058***

0.544***

59 0.057***

0.709***

60 0.050***

0.671***

61 0.045***

0.568***

62 0.045***

1.028***

63 0.043***

1.278***

64 0.040***

1.949***

Coefficient on age group dummies

21

0.006*** -0.066***

22

0.010*** -0.066***

23

0.015*** -0.074***

24

0.020*** -0.077***

25

0.022*** -0.072***

26

0.028*** -0.067***

27

0.027*** -0.075***

28

0.032*** -0.070***

29

0.036*** -0.075***

30

0.029*** -0.076***

31

0.033*** -0.085***

32

0.035*** -0.085***

33

0.033*** -0.077***

34

0.033*** -0.101***

35

0.030*** -0.099***

36

0.029*** -0.114***

37

0.027*** -0.106***

38

0.028*** -0.118***

39

0.025*** -0.117***

40

0.022*** -0.118***

41

0.020*** -0.135***

42

0.023*** -0.132***

43

0.020*** -0.171***

44

0.022*** -0.175***

45

0.021*** -0.188***

46

0.016*** -0.148***

47

0.014*** -0.220***

48

0.011*** -0.195***

49

0.012*** -0.197***

50

0.010*** -0.252***

51

0.012*** -0.312***

52

0.008*** -0.221***

53

0.006** -0.216***

54

0.007*** -0.313***

61

55

0.005** -0.336***

56

0.003 -0.328***

57

0.001 -0.287***

58

0.001 -0.486***

59

0.001 -0.645***

60

-0.005* -0.628***

61

-0.008*** -0.544***

62

-0.007*** -0.992***

63

-0.009*** -1.252***

64

-0.011*** -1.933***

Year FE Y Y Y

Obs 17,554 17,554 17,554

R-square 0.176 0.086 0.238

Note: Standard errors are clustered at the country level, but

not reported to save space. * Significant at 10%; **

significant at 5%; *** significant at 1%.

62

Table 6. Robustness Check: Alternative Entrepreneurship Definitions

Dep. Var. esentr_shd nascent ttl_esentr

Sample All OECD

All OECD

All OECD

specifics (1) (2) (3) (4) (5) (6)

Panel A: Country level regressions

r -0.057 -0.061

-0.032 -0.037

-0.058 -0.069

[0.013]*** [0.013]***

[0.007]*** [0.014]**

[0.012]*** [0.018]***

Country

Characteristics Y Y Y Y Y Y

Year FE Y Y

Y Y

Y Y

Country FE N N

N N

N N

Obs 230 393

230 393

230 393

R-square 0.33 0.56 0.48 0.43 0.32 0.53

Panel B: Country-age level regressions

sa 0.244 0.122

0.123 0.081

0.265 0.158

[0.039]*** [0.062]*

[0.028]*** [0.035]**

[0.046]*** [0.068]**

a -0.274 -0.141

-0.157 -0.108

-0.328 -0.208

[0.040]*** [0.062]**

[0.030]*** [0.036]***

[0.048]*** [0.068]***

Year FE Y Y Y Y Y Y

Country FE Y Y

Y Y

Y Y

Obs 17554 10309

17554 10309

17554 10309

R-square 0.69 0.31 0.58 0.36 0.66 0.39

Note:

The dependent variable in column (1) and (2) is "esentr_shd", which is defined as early-stage entrepreneur

plus those businesses that have died in the last twelve months; the dependent variable in column (3) and (4) is

"nascent", which is nascent businesses that are less than 42 months old but do not pay wages; the dependent

variable in column (5) and (6) is "ttl_esentr", which is defined as nascent plus early-stage entrepreneurs.

In panel A, observations are weighted by the number of individuals who make up each country-age-year cell.

In panel B, observations are weighted by the number of individuals who make up each country-year cell.

Standard errors clustered at the country level are in parentheses.

* Significant at 10%; ** significant at 5%; *** significant at 1%.

63

Table 7. Robustness Check: Entrepreneurship with High Aspiration

Types of

early-stage entre.

Plan to hire more than 5

employees in 10 years

Plan to hire more than 10

employees in 10 years

Involve new

products/services

Sample All OECD

All OECD

All OECD

specifics (1) (2) (3) (4) (5) (6)

Panel A. Country level regressions

r -0.007 -0.01

-0.004 -0.006

-0.005 -0.011

[0.002]*** [0.002]***

[0.001]*** [0.001]***

[0.002]*** [0.004]***

Country

Characteristics Y Y Y Y Y Y

Year FE Y Y

Y Y

Y Y

Country FE N N N N N N

Obs 230 393 230 393 230 393

R-square 0.33 0.23 0.36 0.18 0.33 0.26

Panel B: Country-age level regressions

sa 0.039 0.027

0.022 0.016

0.032 0.024

[0.007]*** [0.013]**

[0.005]*** [0.007]**

[0.006]*** [0.011]**

a -0.049 -0.035

-0.028 -0.02

-0.04 -0.03

[0.007]*** [0.013]**

[0.005]*** [0.008]**

[0.006]*** [0.011]**

Year FE Y Y Y Y Y Y

Country FE Y Y

Y Y

Y Y

Obs 17554 10309

17554 10309

17554 10309

R-square 0.23 0.12 0.15 0.09 0.28 0.2

Note:

In panel A, observations are weighted by the number of individuals who make up each country-age-year

cell. In panel B, observations are weighted by the number of individuals who make up each country-year

cell.

Standard errors clustered at the country level are in parentheses.

* Significant at 10%; ** significant at 5%; *** significant at 1%.

64

Appendix

A Proofs of propositions

Lemma 1. For 1

1

r

ra

e

eF , there exists an ]1,0[ma such that for all

maa , 02

drda

Fd, and

for maa , 02

drda

Fd.

Proof. Note that

])1)(1[()1( 2

2rr

r

ra

rearee

e

drda

Fd

(A1)

Therefore

0]1[)1(

lim2

2

0

re

e

e

drda

Fd r

r

r

a

(A2)

and

0]1[)1(

lim2

2

1

re

e

e

drda

Fd r

r

r

a

(A3)

In addition, the function has a unique root

)1,0()1(

1

r

rr

mer

eeraa

(A4)

Given (A2), (A3), and (A4),33

it must be true that for allmaa , 0

2

drda

Fd and

formaa , 0

2

drda

Fd Q.E.D.

Lemma 2. For 1

1

r

ra

e

eF , and 0<a<1, 0

dr

dF.

Proof. Note that

2)1(

)1()1(

r

rrarra

e

eeeae

dr

dF

(A5)

33 The inequalities in (A2), (A3), and (A4) hold because r>1-e-r and r<er-1.

65

Therefore

0lim0

dr

dF

a

(A6)

0lim1

dr

dF

a

(A7)

Given (A6), (A7), and Lemma 1, dF/dr<0 for aϵ[0,1]. Q.E.D.

Proposition 2 Number of entrepreneurs as a fraction of the workforce decreases as the cohort size

shrinks, i.e., / 0dE dr .

Proof. Define 1G F as the inverse of F on variable a, such that

G(F(a,r),r)≡a (A8)

Totally differentiate this equation with respect to r, we have 1 2 2 0G F G . Hence

2 1 2 0G

G G Fr

(A9)

The inequality is true because 1 0G and

1

2 0F

Fr

.

Replace F(a,r) with s,

1

0( ) ( ( ), ( ( , )))E r E h s q G s r ds (A10)

Given (18) and q’<0, we know / 0dE dr . Q.E.D.

Corollary 3. The number of entrepreneurs as a fraction of the workforce decreases with the

population’s median age.

Proof: The median age, amed, is defined by F(amed, r) = 0.5, or by solving 5.01

1

r

ra

e

e.The solution

is )2

1ln(

1

re

ra , which increases monotonically in r (The function resembles the standard logistic

shape). Since, by proposition 2, 0/ drEd , it follows that d E d amed/ 0

66

Proposition 4. There exists some aM, 0<aM<1 such that

r

rE

r

raE M

),0(),(and

r

rE

r

raE M

),1(

),( under the conditions that the first and second

derivatives of V, H, Q exist and are bounded and that the first derivatives of V and H are greater than

some small constant ε.

Proof. Notice that

21

121

2 '''' F

a

HVPFHVP

ra

E

(A11)

Since 0),0(2 rF and 0),1(2 rF , the value of ra

E

2

is dominated by 121 '' FHVP when a

is closed to 0 or 1. Moreover, as shown by Lemma 1, 0),0(12 rF and 0),1(12 rF , therefore

0),0(2

ra

rE and 0

),1(2

ra

rE (A12)

as long as the first derivatives of V and H are greater than some small constant ε.

Given the continuity of P, V, and H, along with the condition (A12), there exist one34

)1,0(Ma ,

such that 0),0(2

ra

rE, i.e.,

r

rE

r

raE M

),0(),(and

r

rE

r

raE M

),1(

),(. Q.E.D.

34 There might exist more than one such local minimum point.

67

B. GEM data coverage

Table B1. GEM Sample Sizes by country and year 2001-2010

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Algeria - - - - - - - - 2,000 -

Angola - - - - - - - 1,518 - 2,167

Argentina 1,992 1,999 2,004 2,003 2,008 2,007 2,018 2,031 2,008 2,001

Australia 2,072 3,378 2,212 1,991 2,465 2,518 - - - 2,000

Austria - - - - 2,197 - 2,002 - - -

Belgium 2,038 4,057 2,184 3,879 4,047 2,001 2,028 1,997 3,989 2,000

Bolivia - - - - - - - 2,000 - 3,524

Bosnia & Herz. - - - - - - - 2,028 2,000 2,000

Brazil 2,000 2,000 2,000 4,000 2,000 2,000 2,000 2,000 2,000 2,000

Canada 1,939 2,007 2,028 2,004 6,418 2,038 - - - -

Chile - 2,016 1,992 - 1,997 2,007 4,008 2,000 5,000 7,195

China - 2,054 1,607 - 2,109 2,399 2,666 - 3,608 3,677

Colombia - - - - - 2,001 2,102 2,001 2,055 11,029

Costa Rica - - - - - - - - - 2,003

Croatia - 2,001 2,000 2,016 2,000 2,000 2,000 1,996 2,000 2,000

Czech Republic - - - - - 2,001 - - - -

Denmark 2,022 2,009 2,008 2,009 2,010 10,000 2,001 2,012 2,012 1,957

Dominican Rep. - - - - - - 2,081 2,019 2,007 -

Ecuador - - - 2,010 - - - 2,142 2,200 2,077

Egypt - - - - - - - 2,636 - 2,769

Finland 2,001 2,005 2,005 2,000 2,010 2,005 2,005 2,011 2,004 2,006

France 1,991 2,029 2,018 1,953 2,005 1,909 2,005 2,018 2,019 2,012

Germany 7,058 15,041 7,534 7,523 6,577 4,049 - 4,751 6,032 5,552

Ghana - - - - - - - - - 2,447

Greece - - 2,000 2,008 2,000 2,000 2,000 2,000 2,000 2,000

Guatemala - - - - - - - - 2,190 2,285

Economy 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Hong Kong - 2,000 2,000 2,004 - - 2,058 - 2,000 -

Hungary 2,000 2,000 - 2,878 2,878 2,500 1,500 2,001 2,000 2,000

Iceland - 2,000 2,011 2,002 2,002 2,001 2,002 2,002 2,005 2,001

India 2,011 3,047 - - - 1,999 1,662 2,032 - -

Indonesia - - - - - 2,000 - - - -

Iran - - - - - - - 3,124 3,350 3,359

Ireland 1,971 2,000 2,000 1,978 2,000 2,008 2,007 2,001 - 2,000

Israel 1,869 2,004 - 1,933 - - 2,019 2,030 2,073 2,007

Italy 1,973 2,002 2,003 2,945 2,001 1,999 2,000 3,000 3,000 3,000

Jamaica - - - - 2,180 3,669 - 2,407 2,012 2,298

Japan 1,999 1,999 2,000 1,917 2,000 2,000 1,860 2,001 1,600 2,006

Jordan - - - 2,000 - - - - 2,006 -

Kazakhstan - - - - - - 2,000 - - -

Korea 2,008 2,015 - - - - - 2,000 2,000 2,001

Latvia - - - - 1,964 1,958 2,000 2,011 2,003 2,001

Lebanon - - - - - - - - 2,000 -

Macedonia - - - - - - - 2,000 - 2,002

68

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Malaysia - - - - - 2,005 - - 2,002 2,010

Mexico 2,014 1,002 - - 2,011 2,015 - 2,605 - 2,605

Montenegro - - - - - - - - - 2,000

Morocco - - - - - - - - 1,500 -

Netherlands 2,013 3,510 3,505 3,507 3,582 3,535 3,539 3,508 3,003 3,502

New Zealand 1,960 2,000 2,009 1,933 1,003 - - - - -

Norway 2,874 2,036 2,040 2,883 2,015 1,999 1996 2049 2029 2,002

Pakistan - - - - - - - - - 2007

Panama - - - - - - - - 2000 -

Peru - - - 2007 - 1997 2000 2052 2021 2108

Philippines - - - - - 2,000 - - - -

Poland 2000 2000 - 2001 - - - - - -

Portugal 2000 - - 1000 - - 2023 - - 2,002

Puerto Rico - - - - - - 1,998 - - -

Romania - - - - - - 2046 2206 2093 2235

Russia 2012 2,190 - - - 1,894 1,939 1,660 1,695 1,736

Saudi Arabia - - - - - - - - 2,000 2,000

Serbia - - - - - - 2,200 2,297 2,300 -

Singapore 2004 2005 2008 3852 4004 4011 - - - -

Slovenia - 2030 2012 2003 3016 3,008 3020 3019 3030 3012

South Africa 1,827 6,993 3,262 3,252 3,268 3,248 - 3,270 3,135 3,279

Spain 2016 2000 2000 16980 19384 28306 27,880 30,879 28,888 26388

Sweden 2056 2000 2025 26,700 2002 2003 2001 - - 2,492

Switzerland - 2001 2003 - 5456 - 2148 - 2024 2,002

Syria - - - - - - - - 2,002 -

Taiwan - 2,236 - - - - - - - 2,001

Thailand - 1,043 - - 2,000 2,000 2000 - - -

Tonga - - - - - - - - 1184 -

Trinidad & T. - - - - - - - - - 2,016

Tunisia - - - - - - - - 2,000 2,001

Turkey - - - - - 2,417 2400 2400 - 2,401

Uganda - - 1035 2005 - - - - 2095 2267

United Arab Em. - - - - - 2001 2180 - 2056 -

United Kingdom 4899 16002 22010 24006 11203 43033 41829 8000 30003 3000

United States 1,983 7,059 9,197 2,007 2,021 3,093 2,166 5,249 5,002 4,000

Uruguay - - - - - 1997 2000 2,027 2001 2,034

Vanuatu - - - - - - - - - 1,182

Venezuela - - 2,000 - 2,000 - 1,794 - 1,693 -

West Bank & Gaza

Strip

- - - - - - - - 2080 1992

Yemen - - - - - - - - 2,065 -

Zambia - - - - - - - - - 2,039

Total 66602 115770 96712 145189 117833 171,631 155,183 134,990 183,074 176,699

Source: Bosma et al. (2012).